AI Intelligence Hub

DigitalOcean announced on 2026-10-12 a managed data layer for agents (Data Flows, one catalog of records and vectors, SQL/API querying, multimodal ingestion, audit logging). Agent Learning Services adds Insights, Signals, Evaluations and Simulations that use stored agent traces to find failures and test fixes. Both are in preview with no pricing yet.

Meta released Muse Wonder and Muse Wonder Full on 2026-10-12, multimodal models for agentic tasks built around Active Memory and Active Focus. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full is limited to Muse Code subscribers.

Mistral AI released Mistral Large 4 in public preview on 2026-10-06 via its API and Mistral Studio. It is a natively multimodal 1T-parameter mixture-of-experts model with ~49B active parameters and 160+ languages, priced at $1.36 input / $4.18 output per million tokens (USD). Open weights are promised by the end of October 2026.

OpenAI released GPT-6.1 Sol on 2026-09-29 with near-GPT-6 Astra performance on coding, professional document analysis, multi-step business workflows and computer use. API pricing is $2/M input, $0.10/M cached input and $10/M output tokens (model id gpt-6-1-sol). Available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu; an Ultrafast variant with 8x faster generation is coming soon.

NVIDIA announced the Open Agent Safety Platform on 2026-09-28, combining OpenShell (open-source secure runtime enforcing boundaries for agents) with Sentry, an out-of-band watchdog on BlueField-4 DPUs that can quarantine policy-violating agents within milliseconds. Over 100 organizations including Anthropic, Microsoft, Salesforce and SAP are participating, with interoperability under the Open Secure AI Alliance at the Linux Foundation.

Anthropic released Claude Sonnet 5.5 on 2026-09-28, running 30%+ faster than Sonnet 5 and scoring 70.6% on Terminal-Bench 4.0 (vs 10.3% for Sonnet 5). API pricing is unchanged at $2/M input and $10/M output tokens; available on Claude Platform (claude-sonnet-5-5), AWS, Google Cloud and Azure. Claude Haiku 5.5 is slated for the coming weeks.

ElevenLabs released Eleven v4 (expressive text-to-speech and improved voice cloning across 90+ languages) and Eleven v4 Turbo (matching quality at roughly 100 ms latency for real-time use) on 2026-09-28. The release adds a Text to Dialogue API, WebSocket streaming for agents and expanded LLM options in ElevenAgents.

Anthropic released Claude Opus 5.5 on September 22, 2026, rolling it out across the API, AWS, Google Cloud, and Microsoft Azure. The model performs at roughly Claude Fable 5.1's level on most work while costing about 40% less and generating output ~30% faster, scoring 66.4% on Terminal-Bench 4.0 versus Opus 5's 52.3%.

OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, two lower-cost GPT-6-family models rolling out via API, Codex, and ChatGPT. Sol targets complex coding and agentic work with roughly half the mistake rate of its predecessor on internal factuality evals; Luna handles high-volume, low-complexity tasks. Both are priced at $2/$10 per million input/output tokens, half of GPT-6 Astra's rate.

Google open-sourced AX v0.3.0 on September 21, 2026, a Kubernetes-style declarative orchestration runtime for running large numbers of autonomous agent workloads in a cluster. The update splits the runtime into three services and moves task state to Redis Streams to handle high volumes of short-lived agent tasks. It ships under Apache 2.0 on GitHub.

Andon Labs launched Pion, a platform where persistent AI agents operate under an "Andonos" overseer to run day-to-day business operations, including ordering supplies, managing a storefront, and handling routine decisions with minimal human input. It builds on Andon Labs' earlier Vending-Bench agent benchmarks and went live around September 15, 2026.

Perplexity brought its Portable Computer local AI agent to Windows PCs equipped with NVIDIA RTX GPUs (24GB+ VRAM), running multistep tasks entirely on-device instead of in the cloud. It launched September 14, 2026, extending Perplexity's existing Comet browser-agent work to a local, offline-capable form factor.

Meta released Muse, a personal AI agent that executes multi-step tasks rather than only answering questions. It runs on Muse Secure VM, a dedicated secure machine with its own browser, and can open a browser, fill out forms, send email, book travel and negotiate on the user's behalf across connected apps. Muse is powered by Muse Spark and ships with a free tier plus Power ($20/mo) and Maximum ($100/mo) paid tiers.

OpenVDN released VDN-H3, an open video generation model built on the MiniMax-H3 architecture. The release includes the blog write-up, training and inference code, and model weights, with multiple variants offering different inference-speed trade-offs.

SpaceXAI made Grok Bot generally available to enterprises. Bots are persistent AI teammates that each get their own computer, sign into the tools a team already uses, and carry jobs end to end across apps and inboxes. Bots save a workflow as a reusable routine, absorb corrections, and re-run it autonomously; enterprises get access, network and audit controls for governing Bots at scale.

OpenAI released GPT-6 Astra, a frontier model with state-of-the-art results on computer use, browsing, software engineering, cybersecurity and professional work. Astra creates documents, spreadsheets and presentations that follow user templates and adapts when requirements change mid-task. OpenAI reports it completes computer-use tasks roughly 1.9x faster than GPT-5.6 Sol and adds agent-monitoring safeguards that can pause a run for user review.

Meta shipped Muse Spark 1.3 via the Meta Model API, adding a max-reasoning mode for hard reasoning and agentic tasks. The model tracks context and prior results across long runs, works through messy or conflicting inputs, and asks for clarification when needed. It is tuned for long-horizon coding workflows with fewer unnecessary turns and cleaner output.

Google released Gemini 3.8 Flash, its workhorse model, alongside Gemini 3.8 Flash Cyber, a gated cybersecurity variant for vulnerability detection and automated patching available to trusted defenders via the Fairwind Program. 3.8 Flash improves on 3.7 across software engineering, agentic tasks and multi-step reasoning at the same introductory price of $0.75/M input and $3.75/M output tokens, and posts a significant gain in prompt-injection robustness as measured by Gray Swan.

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 — the same underlying model shipped with two different safeguard levels, with Fable 5.1 generally available and Mythos 5.1 restricted to vetted cybersecurity and life-sciences programs. Fable 5.1 scores 55.8% on Terminal-Bench 4.0, 52.6% on Terminal-Bench-Science 0.1, and 73.4% on CursorBench 3.2.0. Cache-read pricing dropped 75% to $0.25 per million tokens, cutting typical workload cost about 25% and highly agentic workload cost up to about 45%; input/output pricing is unchanged at $10/$50 per million.

openKylin 3.0 is an open-source Linux distribution that treats AI agents as native OS residents rather than installed applications, unifying models, memory, tools, system permissions, and desktop capabilities into a single system-level runtime. Agents invoke desktop capabilities through the MCP protocol instead of simulated clicks, and the release explicitly supports third-party agents including OpenClaw, WorkBuddy, and Raccoon Work alongside its own KylinBot. It ships on the Linux 7.0 kernel across x86, ARM, RISC-V, and LoongArch, with a unified AI SDK and an openKylin-skills repository for shared agent skills.

Apple introduced the M6, its first 2nm SoC, with a 12-core CPU, 12-core GPU and a dual 16-core Neural Engine, alongside the M5 Ultra, a quad-die part built by joining two dual-die M5 Max chips over next-generation UltraFusion with up to a 36-core CPU, 80-core GPU and 32-core Neural Engine. Apple claims the M6 delivers the fastest single-threaded performance available and up to 2.4x the CPU performance of the M1. The M6 ships in the new Mac mini and the M5 Ultra in the new Mac Studio.

Google Cloud released Gemini Enterprise for Financial Services in preview, a vertical packaging of its Gemini Enterprise agent platform aimed initially at capital markets and corporate banking. It ships a Google-managed Financial Research agent, more than 50 new skills carrying role- and workflow-specific agentic instructions, enterprise data connectors, and a third-party agent ecosystem.

Apache Maka (Incubating) is a local-first AI agent workspace in which model messages, tool calls, tool results, permission decisions and termination events are all recorded as an append-only log. It inspects projects and runs tools inside a sandbox boundary through a single Runtime Host, with sessions, settings and run data kept on the user's machine by default while model connections can point at cloud APIs, local models or compatible gateways. Desktop, terminal TUI, non-interactive CLI and evaluation interfaces all share the same Runtime Host.

NVIDIA announced that the Groq 3 LPX, a specialised interactive AI inference accelerator, has entered full production. It is positioned as an extension of the NVIDIA Vera Rubin platform and is engineered specifically for ultrafast token generation in agentic systems rather than general training workloads.

Thomson Reuters launched Thomson, its first proprietary large language model, built in-house by post-training Alibaba's open-weight Qwen 3.5 base on Thomson Reuters legal and tax content. The final training run cost roughly USD 450,000 against a two-year total investment of about USD 40 million in talent and compute, and the model was trained on under 10% of the company's content library. It debuts inside Tabular Analysis in CoCounsel Legal, with wider rollout across the legal and tax portfolio and sovereign AI options planned.

First open-weight model in Dots Studio's third-generation dots3 family, released under Apache 2.0. A mixture-of-experts model with roughly 280B total parameters and about 16B active per token, a 512K context window, and multimodal input across text, image, video and audio with text output. Weights are on Hugging Face and ModelScope with code on GitHub.

Z.ai's flagship coding and agent model, built on the same base model as GLM-5.2 with every gain derived from scaled post-training. Z.ai reports a 50% improvement over GLM-5.2 on its Code Bench and a lift on Terminal-Bench 3.0 from 4.6 to 28.3. Supports a 1M-token context, 128K maximum output and always-on reasoning at low, high and max effort; open weights were withheld at launch pending a cyber-safety review.

Apache 2.0 native multimodal dense model published to Hugging Face at 55.6 GB of BF16 safetensors across 18 shards. Offers 262K native context extendable toward 1M tokens via YaRN, using a hybrid layout of Gated DeltaNet and Gated Attention blocks across 64 layers. Alibaba positions it as outperforming Qwen3.7-Plus overall with particular strength in real-world coding and office workflows.

Google's newest Flash-tier Gemini model, shipped 13 August 2026 and available same-day through Google AI Studio and third-party gateways including batch endpoints. It is the speed-and-cost tier of the Gemini 3.x line rather than a frontier reasoning release.

WRITER shipped Palmyra X6 alongside major upgrades to the WRITER Agent harness and new governance tooling for controlling token spend. Palmyra X6 is priced at $2 in / $8 out per million tokens and completes an average task for roughly $0.12, about half the previous generation's per-task cost in about half the time. Paired with the new harness, WRITER Agent reports 52% lower average cost, 48% better speed and 10% higher quality.

MIT-licensed agent harness released in developer preview, built on the Cordis meta-framework where models, tools, skills, sessions, sandboxes, storage, loops and the UI are all swappable plugins. Everything the model sees is written to an append-only session log, inspectable by source in a Trajectory view. Reported to pass 135,000 GitHub stars within four days of release.

Meta released Muse Glimmer on 2026-08-10, a 30B-parameter open-weight agentic model distilled from Muse Spark 1.2 under an Apache 2.0 license. It is optimized to run locally on a single consumer GPU (Mac or PC) and targets multi-step agentic workloads such as coding, web research, and document analysis. It marks Meta's push toward on-device 'personal intelligence' that runs without cloud dependency.

MiniMax open-sourced H3 (Hailuo 3.0) on August 3, 2026, publishing weights to Hugging Face (MiniMaxAI/MiniMax-H3) with same-day native ComfyUI support. The unified multimodal model reads text, image, video, and audio as one context and generates 4–15s clips at 2K/24fps with native stereo audio in a single pass—no separate audio stage or upscaler. It ships in Diffusers format with FL2VA and Ref2VA checkpoints (smallest working combo ~42.5GB) under a community license, ranking #1 in Video Editing and #2 in Text-to-Video on Artificial Analysis boards.

Alibaba's Qwen team made its flagship Qwen3.8-Max broadly accessible in early August 2026 (previewed July 19 at WAIC Shanghai), a 2.4-trillion-parameter mixture-of-experts model with ~95B active parameters per request and a 1M-token context. It targets coding, agentic workflows, multimodal input, and long-horizon 'professional cowork' tasks, posting frontier-competitive scores (86.6 Terminal-Bench 2.1, 93.0 PaperBench). Open weights on Hugging Face and ModelScope were announced for the following week, making it the first Max-class Qwen model slated to be open-sourced.

Liquid AI released two open-weight encoder models — LFM2.5-Encoder-230M and 350M — on July 28, 2026, targeting fast long-context inference on CPU. They support an 8,192-token context, match or beat larger encoders on GLUE/SuperGLUE and multilingual tasks, and run about 3.7x faster than ModernBERT-base at long context.

Kimi K3 is Moonshot AI's 2.8-trillion-parameter sparse Mixture-of-Experts model with native multimodality (text, image, video), a 1M-token context window, and an always-on reasoning mode built on Kimi Delta Attention. Full open weights were published to Hugging Face on July 27, 2026 under the Kimi K3 License, with MXFP4 quantization-aware training reducing the weight footprint to roughly 1.4 TB.

Google DeepMind's new generally-available workhorse model, released July 21, 2026. It improves coding, knowledge work, and multimodal performance while cutting token usage by up to 17% (up to 65% on some DeepSWE agent tasks) versus 3.5 Flash. Supports a 1M-token input context and 64K output tokens, priced at $1.50/$7.50 per million input/output tokens via API.

Poolside's open-weight agentic coding model released July 21, 2026 — a 118B-parameter Mixture-of-Experts (~8B active per token) with a 1M-token context. On Terminal-Bench 2.1 (70.2%) and SWE-Bench Pro it matches or beats models many times larger, and is small enough to run on a single NVIDIA DGX Spark. Weights are on Hugging Face under an OpenMDW-1.1 license, with API access and OpenRouter availability.

A cybersecurity-specialized model built on Gemini 3.5 Flash, announced July 21, 2026, fine-tuned to find, validate, and patch software vulnerabilities cheaply. Running inside Google's CodeMender agent, it found more unique confirmed vulnerabilities than mainline 3.5 Flash and Claude Opus 4.6 in testing (55 vs 47 vs 36 on the V8 JavaScript engine). Available via a limited-access pilot to governments and trusted partners.

OpenAI's GPT-5.6 family reached general availability on 2026-07-09 in three tiers: Sol (flagship, tuned for coding, biology, chemistry and cybersecurity), Terra (balanced everyday model), and Luna (most cost-efficient). It is now ChatGPT's default, with pricing spanning roughly $1/$6 (Luna) to $5/$30 (Sol) per million input/output tokens.

Meta shipped Muse Spark 1.1 on 2026-07-09 as its first paid model (about $1.25/$4.25 per million input/output tokens), leaning into agentic work, computer use, and a 1,000,000-token context window.

ByteDance's flagship text-to-image and image-editing model, positioned above the 5.0 and 5.0 Lite tiers. It splits a single render into 10+ editable PNG layers, reads spatial annotations, and renders in-image text across roughly 14 languages. Available via Volcano Ark/Engine, BytePlus internationally, the Doubao and Jimeng apps, and third-party hosts like fal and ComfyUI.

OpenAI's new full-duplex voice model powering ChatGPT Voice, able to listen and speak simultaneously with natural backchanneling ('mhmm', 'yeah') instead of turn-taking. GPT-Live-1 is the default for Go/Plus/Pro users and GPT-Live-1 mini for Free users, replacing Advanced Voice Mode. It runs on GPT-5.5 in the background and delegates harder tasks to a frontier model mid-conversation.

Anthropic's new mid-tier model, released June 30, 2026, positioned as its most agentic Sonnet yet — it can plan, use tools like browsers and terminals, and run autonomously at a level that recently required larger models. Performance is close to Opus 4.8 at much lower cost. It is the default model for Free and Pro plans and available to Max, Team, and Enterprise, with introductory API pricing of $2/$10 per M tokens through Aug 31, 2026 (then $3/$15).

Launched June 29, 2026, ClinePass is a low-cost subscription ($4.99 first month, then $9.99/mo) that bundles curated open-weight coding models — GLM-5.2, Kimi K2.7 Code, DeepSeek V4 Pro/Flash, MiniMax M3, MiMo V2.5, Qwen3.7 — with Cline's agent harness and 2–5x standard rate limits across every surface Cline runs on, with no separate provider accounts or API keys.

Claude Tag adds Claude to Slack as a persistent, shared AI teammate: anyone in a channel can @Claude to delegate a task, and Claude works through it asynchronously in stages, posts results back to the thread, and builds memory of the team's work over time. It supports an ambient mode that proactively surfaces relevant information, and launched June 23, 2026 in beta for Claude Enterprise and Team customers. The older 'Claude in Slack' app is being retired on August 3, 2026.

Mistral OCR 4 is a compact document-understanding model that returns extracted text plus bounding boxes, typed-block classification (titles, tables, equations, signatures) and per-word confidence scores across 170 languages. It runs in a single container for fully self-hosted deployment and acts as an ingestion layer for RAG and enterprise search. Priced at $4 per 1,000 pages via API ($2 batch).

On June 22, 2026 OpenAI shipped the full GPT-5.5-Cyber, its strongest model for finding and patching software vulnerabilities (85.6% on CyberGym), alongside an updated Codex Security plugin that automates discovery and patching of flaws and helps prevent new ones from reaching production. The release is part of OpenAI's Daybreak cybersecurity program, with access limited to verified defenders under added monitoring and controls, plus a 'Patch the Planet' open-source remediation initiative.

Krea released open-weights checkpoints of its Krea 2 text-to-image foundation model — a 12B-parameter diffusion transformer — in two variants: Krea 2 Raw (base) and Krea 2 Turbo (distilled for fast few-step inference). Weights are published on Hugging Face for self-hosting and fine-tuning.

GLM-5.2 is Zhipu AI's (Z.ai) open-weights mixture-of-experts LLM — 744B total / 40B active parameters with a 1M-token context window, optimized for agentic coding. It rolled out to paying coding customers on June 13, 2026, with the MIT-licensed weights released on Hugging Face (zai-org) on June 17, 2026. Independent measurements rank it the most capable openly available model (Intelligence Index v4.1 score ~51), ahead of MiniMax-M3, DeepSeek V4 and Kimi K2.6.

Hugging Face released Serge on 2026-06-12 — an open-source (Apache-2.0), GitHub-native AI code-review system that reviews pull requests using any OpenAI-compatible model endpoint (OpenAI, the Hugging Face Router, local vLLM/TGI/LM Studio, or custom providers). Maintainers trigger it on demand and define repository-owned review policies, keeping model choice under the team's control.

Moonshot AI launched Kimi Work (reported 2026-06-12), a local desktop AI agent for macOS (Apple Silicon) and Windows that automates work across files, browsers and code execution, reportedly running on Kimi K2.6. It uses an agent-swarm architecture of up to 300 parallel local sub-agents, a browser-automation extension, a scheduling engine, persistent memory, and native market-data access. It is in early/internal testing.

Cohere open-sourced North Mini Code on 2026-06-09 under Apache 2.0 — a 30B mixture-of-experts agentic coding model (3B active per token) that runs on a single NVIDIA H100 at FP8 with 256K context. Built for agentic software engineering including sub-agent orchestration, architecture mapping, code review and terminal work, with weights on Hugging Face.

Anthropic released Claude Fable 5 on 2026-06-09 — a publicly accessible, safety-constrained version of its frontier Mythos-class model, state-of-the-art across software engineering, knowledge work, vision and science (~95% SWE-bench Verified, ~80% SWE-bench Pro). Within roughly a day it was temporarily taken offline following a U.S. government export directive and is not currently publicly available. Pricing was $10/$50 per million input/output tokens.

At WWDC 2026 (2026-06-08) Apple shipped its third-generation on-device Foundation Models (AFM 3), headlined by a 20B-parameter sparse model (AFM 3 Core Advanced) that activates only 1-4B parameters per prompt via Instruction-Following Pruning and now accepts image inputs. Apple also introduced Core AI, a lower-level framework replacing Core ML for running arbitrary on-device models such as Qwen, Mistral and SAM3, with PyTorch-to-Apple-Silicon conversion tools. The Foundation Models framework and many new APIs are going open source, and on-device plus Private Cloud Compute access is free for developers under 2M first-time App Store downloads.

Apple's rebuilt Siri AI moved from developer beta to general rollout in September 2026, bringing personal context awareness, on-screen content detection, and cross-app actions to iOS 27, iPadOS 27, and macOS 27. The assistant runs on Apple's Private Cloud Compute with a user-selectable underlying model (ChatGPT, Gemini, or Claude).

OutSystems unveiled its open Agentic Systems Platform on 2026-06-03 at the ONE Conference, powered by the OutSystems Enterprise Context Graph. It adds Agent Experience (a suite of A2A and MCP tools and services), Agentic Enterprise Orchestration with guardrails and agent evaluations, a native Kiro (AWS) integration, AWS Transform legacy-modernization pipelines, and a first Banking loan-origination industry solution.

MAI-Thinking-1, unveiled at Microsoft Build 2026 on 2026-06-02, is Microsoft's first in-house reasoning model — a mid-sized sparse MoE with 35B active parameters and a 256K-token context window, trained entirely on commercially licensed data with no distillation from third-party models. It reaches 97.0% on AIME 2025 and 94.5% on AIME 2026 and matches Claude Opus 4.6 on SWE-Bench Pro.

Microsoft launched MAI-Code-1-Flash at Build 2026 on 2026-06-02 — its first in-house, Copilot-native coding model, built end-to-end on clean, appropriately licensed data. It rolled out to all GitHub Copilot plans for fast, efficient everyday coding assistance and reports better price-to-performance than Claude Haiku 4.5 across coding benchmarks.

Released 2026-06-02 on HuggingFace, Holo3.1 is H Company's family of vision-language computer-use agents (0.8B / 4B / 9B / 35B-A3B), built on the Qwen base. It adds robustness across web, desktop, and mobile, native function-calling for agent frameworks, and — for the first time in this family — quantized checkpoints (FP8, Q4 GGUF, NVFP4) tuned for local inference. The 35B model scores 79.3% on AndroidWorld; smaller variants jump from 58% to 72%.

MiniMax M3 (API live since 2026-06-01) began its open-weights rollout on Hugging Face/GitHub around 2026-06-10 to 06-11, alongside its technical report; independent confirmation of the public checkpoint was still settling as of 2026-06-12. M3 pairs frontier agentic-coding performance, a 1M-token context window, and native multimodality (image/video input plus desktop computer use) via its MSA (MiniMax Sparse Attention) architecture, topping the open-weight SWE-Bench Pro at 59.0% — ahead of GPT-5.5 and Gemini 3.1 Pro — at roughly one-tenth the input cost.

NVIDIA moved the Vera Rubin NVL72 rack-scale system from announcement into shipping form on 2026-08-24, positioned as next-generation infrastructure for agentic AI workloads and claiming up to 30x more work per watt. NVIDIA cites OpenRouter data showing agentic workloads consume up to 15x more tokens than standard chat requests. The platform is simultaneously being extended with the Groq 3 LPX interactive inference accelerator, now in full production.

New Claude Code capability launched May 28, 2026 in research preview. Claude writes a JavaScript orchestration script that runs up to 16 concurrent subagents and up to 1,000 total subagents per session, with verification and convergence built in. Available across the Claude Code CLI, Desktop app, VS Code extension, the Claude API, and Amazon Bedrock for Max, Team, and Enterprise plans.

Anthropic's new flagship Opus model, released May 28, 2026, with sharper judgement, more honest progress reporting, and longer independent execution than predecessors. Available immediately on the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and GitHub Copilot. Adds high-effort default mode, restored tool-calling reliability, and a cheaper Fast mode.

Gemini Spark is Google's 24/7 AI agent that runs on Google Cloud VMs, persisting even when the user's device is closed or locked. It operates autonomously across Gmail, Google Sheets, Docs, Drive, and Calendar, surfacing progress via Android Halo notifications. Launched for Google AI Ultra subscribers in the US ($100/month) on approximately May 26, 2026, following its announcement at Google I/O on May 19.

xAI's Grok Build 0.1 is an agentic coding CLI and API purpose-built for interactive, multi-step software engineering. The model (grok-build-0.1) features a 256K context window, native parallel sub-agents, ACP support, headless/scripting mode, and deep Git worktree integration. The API went live May 20, 2026; the consumer beta opened May 25 to all SuperGrok and X Premium Plus subscribers. Priced at $1/M input and $2/M output tokens.

Microsoft announced general availability of computer-use agents in Copilot Studio on May 22, 2026. Enterprise agents can now interact directly with software interfaces, using visual reasoning to navigate websites, forms, and legacy business systems dynamically — without API integration. Ships with Agent 365 governance, audit logging, and MCP tool support.

Google's new Flash-tier frontier model, announced at I/O 2026 on May 19 and generally available May 20, 2026 across the Gemini app, AI Mode in Search, Antigravity, the Gemini API in AI Studio and Android Studio, and the Gemini Enterprise Agent Platform. Outperforms Gemini 3.1 Pro on agentic and coding benchmarks while running ~4x faster on output tokens per second than other frontier models.

Antigravity 2.0 is Google's standalone agent-first development platform — desktop IDE, Go-based CLI, SDK for custom agents, Managed Agents tier on the Gemini API, and an enterprise deployment path, powered by Gemini 3.5 Flash. The Antigravity CLI went generally available and the legacy Gemini CLI was officially sunset on June 18, 2026 for AI Pro, Ultra and Code Assist users, making Antigravity the sole Google terminal agent. It runs parallel agents, custom subagents, scheduled background tasks, and native voice commands.

Gemini Omni is Google's new world model that combines text, image, audio, and video inputs in a single prompt to produce cinematic-quality video with physics-aware reasoning. Omni Flash launched May 19, 2026 with 10-second clip caps and immediate availability to Google AI Plus, Pro, and Ultra subscribers via the Gemini app and Google Flow.

Cursor shipped Composer 2.5 on May 18, 2026 — a Kimi K2.5-based agentic coding model tuned for long-horizon sessions, priced at $0.50 per million input tokens, with benchmark scores rivaling Claude Opus 4.7 on SWE-Bench and Terminal-Bench. Designed for sustained agentic loops at ~200 tok/s.

xAI shipped persistent expertise for Grok on May 18, 2026 — generating documents, decks, and spreadsheets, automating workflows, and letting users build and share custom skills — and made SuperGrok and X Premium subscriptions usable inside OpenClaw, the open-source local-first agent assistant, on May 19, 2026.

Qwen3-Coder-Next is an 80B-parameter open-weight coding model from Alibaba's Qwen team using an ultra-sparse MoE design (~3B active per forward pass), released under Apache 2.0 on May 18, 2026 via LLM Gateway and Hugging Face. Engineered for repo-scale reasoning and agentic coding rather than autocomplete.

Microsoft turned Copilot Studio into an AI agent control center on May 14, 2026, adding agent posture surfacing in the authoring UI, Analytics Viewer role GA, MCP server tool support in workflows (preview), and Agent 365 integration as the control plane for observing, governing, and securing AI agents across Microsoft and ecosystem partners. Admins can now inspect authentication gaps and policy impacts at the source.

On May 14, 2026, OpenAI brought Codex control into the ChatGPT mobile app on iOS and Android, letting developers monitor, approve, and steer Codex sessions running on remote laptops, devboxes, or SSH-connected servers. Remote SSH support also graduated from preview to GA. Files, credentials, and local setup stay on the host machine; phone receives real-time screenshots, terminal output, diffs, and test results.

Amazon launched Alexa for Shopping on May 14, 2026, replacing the 2024 Rufus assistant across mobile, desktop, and Echo Show. Powered by Alexa+, the agent compares products, tracks prices, schedules recurring orders, and uses a Buy for Me feature to purchase items from non-Amazon retailers on the user's behalf. Personalised via user habits and purchase history; available to all US customers without a Prime or Echo requirement.

Microsoft pushed Edge 136 on 2026-05-14 with a Copilot update that reads and reasons across all open tabs simultaneously — compare products, summarize articles, and pull context from history under user permission. Replaces last year's Copilot Mode with multi-tab analysis, AI-generated podcasts, Journeys-on-mobile, study tools, and stricter privacy controls. Automatic rollout completes by 2026-05-27.

OpenAI shipped a new generation of voice intelligence models in the API on 2026-05-14, advancing real-time speech-to-speech latency, multilingual speaker handling, expressive prosody, and developer controls for building voice agents. Positioned as the API foundation for next-generation phone, support, and embedded voice agents.

Anthropic launched Claude for Small Business inside Claude Cowork on May 13, 2026, bundling 15 agentic workflows and 15 reusable skills with connectors to QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, Microsoft 365, and Slack. Workflows target month-end close, payroll planning, invoice chasing, lead triage, contract review, and campaign creation. Available on Team or Enterprise plans with human-in-the-loop approval before any send, post, or payment.

Notion launched its Developer Platform on 2026-05-13, introducing Workers (hosted code runtime in a secure sandbox, deployed via CLI), an External Agent API with launch partners Claude Code, Cursor, Codex, and Decagon, two-way webhooks, and database sync to keep records fresh from Zendesk, Salesforce, Postgres, and other systems. Free during beta; Workers move to Notion credits on 2026-08-11.

Cursor 3.1/3.2 shipped in May 2026 with Canvases — durable, interactive side-panel artifacts (dashboards, tables, diagrams, diffs, to-do lists) that let agents render visual output instead of dense text — and Agent Development Environments with multi-repo support, Dockerfile-based config, faster caching, build secrets, version history, audit logs, rollback, and scoped secrets.

Anthropic shipped Agent View in Claude Code 2.1.139 on 2026-05-11 as a Research Preview CLI dashboard that turns parallel Claude Code sessions into a single command-line control center. Developers open one screen, kick off agents with /bg or claude --bg [task], see which sessions are running, blocked, or done, and jump in only when input is needed. Available across Pro, Max, Team, Enterprise, and Claude API plans by running claude agents.

On 2026-05-11 Google rolled new AI generation into Google Vids, powered by Lyria 3 (music) and Veo 3.1 (video). High-quality video generation is now available at no additional cost inside Workspace, with 4K output, native 9:16 vertical support, and enhanced character consistency from the underlying Veo 3.1 platform.

OpenAI's flagship real-time voice model, GA on May 8, 2026, with GPT-5-class reasoning, a 128K token context window (up from 32K), and audio I/O via the Realtime API. Scores ~15% higher than GPT-Realtime-1.5 on Big Bench Audio and ~14% higher on Audio MultiChallenge at high effort. Priced at $32/M audio-input and $64/M audio-output tokens, with $0.40/M cached input.

Real-time voice translation model launched alongside GPT-Realtime-2 in OpenAI's Realtime API on May 8, 2026. Translates live voice conversations across 70+ input languages and 13 output languages with conversational pacing. Priced at $0.034 per minute.

Paper posted to arXiv on May 7, 2026 by Daniel Zheng et al. (Google DeepMind). A stateful, asynchronous multi-agent workbench built on Gemini 3.1 Pro with a project-coordinator agent dispatching sub-agents across parallel research workstreams, tracking failed hypotheses and producing LaTeX write-ups with provenance. Used by Oxford mathematician Marc Lackenby to resolve open problem 21.10 from the Kourovka Notebook.

Cofounder 2 is an agentic operating system that orchestrates AI agents across engineering, sales, marketing, ops and design — framed as 'infrastructure for the one-person billion-dollar company.' Uses a superoptimizer manager agent that delegates work to departmental sub-agents, with a layered working / core / long-term memory model. Includes a fellowship program offering $1,000 upfront capital and $100/day in platform credits.

Released May 6, 2026 by Zyphra. A reasoning Mixture-of-Experts language model with 760M active / 8.4B total parameters, trained end-to-end on AMD Instinct MI300X GPUs with AMD Pensando Pollara networking on IBM Cloud. Introduces Compressed Convolutional Attention (CCA), an MLP-based expert router and learned residual scaling. Open weights on Hugging Face under Apache 2.0; free serverless endpoint on Zyphra Cloud.

Released May 5, 2026 at Anthropic's invite-only financial services briefing in New York. Ten ready-to-run Claude agent templates targeting banking, insurance, asset management and fintech — covering pitchbook generation, KYC screening, compliance review, financial-statement analysis and month-end close. Pairs with Claude Opus 4.7 and Claude add-ins for Microsoft Excel, PowerPoint, Word and Outlook.

Microsoft's enterprise control plane for AI agents went GA on May 1, 2026 at $15/user/month, and the May 2026 capabilities update (early May) added a Shadow AI page in Defender/Intune to detect locally running agents on Windows endpoints, plus public-preview registry sync that automatically discovers and inventories agents from AWS Bedrock and Google Cloud (Gemini Enterprise) under one pane of glass. Bundled into the new Microsoft 365 E7 'Frontier Suite' SKU at $99/user/month alongside E5, Copilot, and Entra Suite.

An open agent-provisioning protocol announced April 30, 2026 that lets AI agents create Cloudflare accounts, start paid subscriptions, register domains, and deploy applications without a human in the loop. Built on three primitives — Discovery (catalog), Authorization (identity attestation via Stripe), and Payment (a scoped Stripe payment token capped by default at $100/month). Launched as part of Stripe Projects (open beta), with Planetscale as a launch partner.

AWS launched Amazon Quick on April 29, 2026 at the What's Next with AWS event — a desktop AI assistant that runs locally on user laptops, connects to Microsoft 365, Google Workspace, Slack, Zoom, Airtable, Dropbox, and Salesforce, and can build custom intelligent apps, dashboards, and web pages from natural-language prompts. Quick proactively surfaces calendar-relevant notes, drafts email replies, and assembles presentations across enterprise apps.

Multiverse Computing released the LittleLamb model family on April 28, 2026 — three ultra-compact ~0.3B-parameter open-source models (LittleLamb 0.3B, 0.3B Tool-Calling, 0.3B Mobile) compressed from Qwen3-0.6B using their CompactifAI tensor-network technology. All three are MIT-licensed on Hugging Face with bilingual English/Spanish support and dual reasoning/low-latency inference modes.

Anthropic shipped 'Claude for Creative Work' on April 28, 2026 — a coordinated launch of nine MCP-style connectors that let Claude operate inside Adobe Creative Cloud (Photoshop, Premiere, Lightroom, Express, Illustrator, InDesign, Stock, Firefly), Blender, SketchUp, Autodesk Fusion, Ableton, Splice, Resolume Arena, Resolume Wire, and Affinity by Canva. Users describe an outcome in natural language and Claude orchestrates multi-step workflows across the relevant pro tools.

xAI rolled out shareable custom templates for Grok Imagine on the web on April 28, 2026, including Photo→Video, Photo→Style Edit, and Photo→Edit→Video chained options. Each template carries a name, description, and shareable link in a dedicated tab, and a fourth Image Reference Edit type is in development with @-mention syntax for swapping reference images mid-prompt.

DeepSeek's open-source flagship MoE family released April 24, 2026: V4-Pro (1.6T total / 49B activated parameters) and V4-Flash (284B / 13B activated), both with native 1M-token context and Apache 2.0 weights on HuggingFace. API supports OpenAI ChatCompletions and Anthropic API formats.

Tencent's flagship 'fast-and-slow-thinking fused' MoE model released April 23, 2026 — 295B total / 21B activated parameters, 256K context, open-sourced on GitHub and HuggingFace. Already integrated across Yuanbao, CodeBuddy, WorkBuddy, Tencent Docs, ima and Peacekeeper Elite.

OpenAI's smartest, most agentic GPT — released April 23, 2026 to ChatGPT Plus/Pro/Business/Enterprise and Codex; expanded to API on April 24 with a 1M-token context window. On May 5, 2026, GPT-5.5 Instant rolled out to all ChatGPT users as the new default model, replacing GPT-5.3 Instant; in API as chat-latest. GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in medicine, law and finance.

Hosted Claude Platform service for long-horizon agent work. At Code with Claude London (May 19, 2026), Anthropic shipped two major enterprise security features: Self-Hosted Sandboxes (public beta) move tool execution into customer-controlled infrastructure via providers including Cloudflare, Daytona, Modal, and Vercel; MCP Tunnels (research preview) connect agents to private internal databases and APIs via outbound encrypted channels, eliminating the need to expose inbound firewall rules. Earlier features from Code with Claude 2026 (May 6–7) remain active: Outcomes, Multiagent Orchestration, Memory, and the Dreams research preview.

PolyAI launched its Agent Development Kit on April 22, 2026, a framework for teams to build, deploy, and iterate AI-native customer-experience agents. Positions PolyAI against Google, OpenAI, and Sierra in the CX agent space with a CX-native rather than general-purpose design.

Google's consolidated enterprise agent platform announced at Cloud Next '26 on April 22, 2026. Evolves Vertex AI into a full agent stack with Agent Studio (low-code), upgraded Agent Development Kit (code-first), Model Garden (200+ models incl. Gemini 3.1, Claude, Gemma 4), Agent Identity, Agentic Defense, and the Agentic Data Cloud. GA on launch day.

OpenAI launched Codex Labs on April 21, 2026 — a hands-on workshop and adoption program that pairs enterprises with top GSIs (Accenture, Capgemini, Cognizant, Infosys, PwC, TCS) to accelerate Codex deployment. Targets Fortune-1000 dev-tool rollouts and AI-native engineering transformations.

Hugging Face released ml-intern on April 21, 2026 — an open-source AI agent that automates the full LLM post-training workflow (SFT data curation, reward modeling, RLHF, evals). Released alongside the Spring 2026 State of Open Source on Hugging Face report.

OpenAI announced gpt-image-2 on April 21, 2026 — its next-generation native image model with reasoning-based generation, 2K resolution output, and multi-image consistency (characters and product IDs persist across a session). Announced at a livestream at 12pm PT on April 21.

Efficiency-focused Gemini variant delivering 2.5× faster response times and 45% faster output generation versus earlier Gemini versions. Priced at $0.25 per million input tokens, making it one of the cheapest frontier-adjacent models available via API.

5N6 unveiled a major update to LiberaGPT enabling a record-breaking 24-billion-parameter large language model that runs entirely offline on the latest-generation iPhone, with no cloud data centers or extra hardware required. Complete privacy with on-device inference.

Microsoft Research published an analysis exploring the intersection of artificial intelligence and global environmental sustainability — examining whether AI's productivity gains can outweigh its own rising energy footprint, and mapping concrete applications in climate modeling, grid optimization, and materials discovery.

MIT Technology Review published a new annual list called '10 Things That Matter in AI Right Now' unveiled at EmTech AI 2026. The list identifies the most significant AI trends and advances MIT reporters will track throughout 2026.

Unified framework for multi-agent multi-view world modeling. Achieves accurate multi-agent control while maintaining multi-view consistency — a long-standing open problem in embodied AI and world-model research.

Google Research unveiled TurboQuant at ICLR 2026, an algorithm that sharply reduces the memory overhead of KV cache in large language models. A two-step pipeline combining PolarQuant vector rotation and a Quantized Johnson–Lindenstrauss compression step cuts memory needs by roughly 6x, making long-context inference dramatically cheaper.

Tufts University researchers developed a neuro-symbolic AI system that cuts energy consumption by roughly 100x versus comparable neural approaches. The hybrid architecture combines symbolic reasoning with neural networks to dramatically reduce the compute needed for equivalent performance on reasoning tasks.

New experimental Anthropic product that lets users create visuals — prototypes, slides, one-pagers, and more — using Claude. Positions Claude as a rapid visual-production agent alongside Claude Code and Claude Cowork.

Microsoft made xAI's Grok 4.1 Fast available in Copilot Studio (preview, US-only, off by default). Makers can now build Copilot agents backed by Grok alongside OpenAI and other models. Elon Musk signalled Grok 4.2 is coming soon.

xAI launched its Grok Speech APIs, bringing real-time voice synthesis and speech understanding to developers through the Grok platform. Rolled out alongside XChat (xAI's WeChat-style messaging app) which uses Grok as its underlying AI engine.

Unveiled at TDX 2026, Headless 360 exposes every capability in the Salesforce platform — CRM, service, marketing, and commerce — as an API, MCP tool, or CLI command, so agents like Claude Code, Cursor, and Codex can build and operate directly on the platform.

Runway's latest video generation model, offering the most granular creative control in the current video-AI market. Supports text-to-video, image-to-video, and AI editing in one tool, with a heavy emphasis on cinematic quality and directable camera/scene control.

OpenAI introduced GPT-Rosalind, a research-preview life-sciences reasoning model for biology, drug discovery, and translational medicine. Optimized for scientific workflows combining improved tool use with deeper understanding across chemistry, protein engineering, and genomics.

Open-source agent from Nous Research that adapts to the user over time. On 2026-06-15 Nous shipped the async_delegation toolset, letting background subagents run as in-process threads that return a task ID immediately, so delegated work no longer blocks the parent chat. The project has surpassed 193K GitHub stars since its February 2026 launch and leads OpenRouter's agent rankings.

Production-ready 1.0 release of Microsoft's unified agent framework for .NET and Python. Combines Semantic Kernel and AutoGen into a single open-source stack with stable APIs, long-term support, full MCP support built-in, and a browser-based DevUI that visualizes agent execution and tool calls in real time.

Open-source AI coding agent framework built in Python and Rust, positioned as an independent foundation for AI-assisted software development. Surpassed 72,000 GitHub stars and 72,600 forks within days of public launch.

Released April 2, 2026 under Apache 2.0 — Google's most capable open-weight model family. Delivers 'frontier-level intelligence per parameter', built for advanced reasoning and agentic workflows. Fully commercially usable with no restrictions.

OpenAI's GPT-5.4 Thinking variant integrates test-time compute and is the first general-purpose model with native computer-use capabilities. Scored 75.0% on OSWorld-Verified — 27.7 points above its predecessor — officially surpassing human-level performance on desktop task benchmarks.

Open-source enterprise agent toolkit from NVIDIA enabling autonomous, self-evolving AI agents with safety, security, and efficiency features.

OpenAI's first general-purpose model with native, state-of-the-art computer-use capabilities built in.

Google's 2026 flagship with adjustable Thinking Mode for deep reasoning and Gemini 3 Flash as its fast companion.

Llama 4 Scout and Maverick are the first Llama models with Mixture-of-Experts (MoE) architecture and native multimodal training. Scout: 17B active params, 16 experts, 109B total, 10-million-token context window — the longest context of any open-weight model.

Meta's 2026 flagship open model with a 10-million token context window — the largest of any publicly available model.

Anthropic's most capable 2026 flagship model with 1M-token context window in beta and best-in-class agentic performance.

AMD's 2025 flagship AI chip with significantly improved bandwidth, competing directly with NVIDIA Blackwell.

Google's open-source terminal agent powered by Gemini. Officially sunset on June 18, 2026 for AI Pro, Ultra and Gemini Code Assist users, replaced by the Antigravity CLI (part of Google Antigravity 2.0). Existing Gemini CLI scripts and pipelines must migrate to the Antigravity CLI command set.

Anthropic's desktop agentic AI that can automate file management, multi-step workflows, and business tasks directly on your computer.

OpenAI's most capable model combining o-series reasoning with GPT-series fluency into a single unified system.

OpenAI's advanced reasoning model — highest-performing model on ARC-AGI at launch, near human-level on many complex benchmarks.

Open protocol by Google enabling independent AI agents to communicate, collaborate, and coordinate across platforms.

OpenAI's open-source terminal coding agent with sandboxed execution, multi-file editing, and git integration.

Google's 2025 flagship with Thinking Mode, achieving industry-best performance across most academic benchmarks at launch.

Multimodal Gemma family in 1B, 4B, 12B and 27B sizes — first Gemma with vision understanding.

Autonomous general-purpose AI agent, now part of Meta. Manus 2.0 (announced 2026-09-28) adds a rebuilt Cascade engine (23.2% fewer tokens, 28.2% faster, 32% cheaper), Manus Studio desktop workspace, Video Editor, Game Dev, Cloud Computer, phone-to-desktop Remote Control, Automations and Cue multi-agent personal agents with their own identities and email addresses.

Apple's 2025 chip delivering 4x the AI performance of M4 with Neural Accelerators in every GPU core.

NVIDIA's Blackwell architecture pairing Grace CPU and B200 GPU — flagship for next-gen AI training and inference clusters.

xAI's third-generation flagship model with real-time X/Twitter data, DeepSearch, and Think mode for extended reasoning.

Anthropic's agentic coding CLI that autonomously reads, edits, and ships code across entire codebases from the terminal.

OpenAI's computer-use agent that autonomously browses the web, fills forms, books reservations, and completes multi-step tasks.

Chinese open-source reasoning model that matched o1 performance at a fraction of the training cost, shocking the AI industry.

Google DeepMind's advanced video generation model — generates high-fidelity, cinematically accurate videos with strong prompt adherence.

AI-powered IDE with Cascade — an agentic coding system that understands full project context and executes multi-step changes.

H100 upgrade with 141GB HBM3e memory and 4.8 TB/s bandwidth — optimized for inference at scale.

Adobe's AI video generation model integrated into Premiere Pro and standalone Firefly platform — designed for commercial-safe creative workflows.

Browser-based AI app builder that generates, runs, and deploys full-stack web applications entirely in the browser.

Prompt-to-fullstack web app builder — generate React + Supabase apps from plain English in minutes.

Alibaba's open-source LLM family — Qwen 2.5 trained on 18T tokens, Qwen 3 supports 201 languages and MoE architecture.

OpenAI's first reasoning model that 'thinks before answering' using chain-of-thought reasoning — excels at math, coding, and science.

Cline shipped Cline Desktop, a native macOS/Windows app that gives its open-source coding agent a dedicated interface with workspace and branch views, moving it beyond a VS Code-only workflow. The app bundles free access to open-weight models (DeepSeek-V4.1-Flash, Musespark-1.3) via ClinePass, plus bring-your-own-key support for any provider, and shipped around September 14, 2026.

Second-generation Gemma in 2B, 9B, and 27B — significantly improved quality, used as default in Google AI Studio free tier.

Anthropic's breakthrough update introducing computer use capabilities — first Claude model that can control a computer.

Fast, high-quality text-to-video and image-to-video AI model producing realistic motion and consistent characters.

Chinese AI video generation model by Kuaishou — produces realistic 2-minute 1080p videos with strong physics simulation.

Omni model combining text, voice, and vision in real-time with dramatically lower latency than GPT-4.

Apple's 2024 chip with dedicated AI accelerators, powering Apple Intelligence features across Mac, iPhone, and iPad.

Google DeepMind's text-to-image model — generates photorealistic images with strong text rendering, available in Gemini apps and Vertex AI.

AWS's AI coding assistant with deep AWS integration — generates, debugs, and transforms code with cloud-native context.

Microsoft's 'small but mighty' model family — Phi-3 (3B/14B) and Phi-4 (14B) outperform models 5x their size on reasoning tasks.

Meta's open-source 8B and 70B models outperforming many closed models — available for free commercial use.

Cohere's enterprise LLM optimized for Retrieval Augmented Generation and tool use — 128K context, multilingual, designed for business document workflows.

AI music generation tool from ex-Google DeepMind researchers — produces full tracks with high audio fidelity and genre control.

Open-source autonomous software engineering agent that browses, codes, runs terminal commands, and resolves GitHub issues.

World's first fully autonomous AI software engineer — plans, codes, tests, and deploys in a sandboxed cloud environment.

Enterprise-focused LLM family optimized for RAG and tool use — Command R+ for accuracy, Command A (2025) runs on just 2 GPUs with 256K context.

Three-tier model family (Opus/Sonnet/Haiku) offering best-in-class performance at different speed/cost trade-offs.

AI-powered cinematic video generation and editing app focused on social-media-ready content and character consistency.

Mistral's flagship commercial model — strong multilingual reasoning, function calling, and 128K context window.

Google's first open-source lightweight model family in 2B and 7B sizes, built from the same research as Gemini.

LPU (Language Processing Unit) chip and cloud API delivering inference at 500+ tokens/second — 10-25x faster than GPU inference.

Text-to-video AI model. OpenAI announced Sora's wind-down: Sora.com web and the Sora app are discontinued April 26, 2026; the Sora 2 API will be discontinued September 24, 2026. Active users dropped from 1M to under 500K and the rumored Disney content deal was never signed.

1 million token context window model — could process entire codebases, long videos, or thousands of documents in one shot.

Stateful multi-actor application framework built on LangChain for building cyclic, graph-based agent workflows.

Perplexity's search-grounded LLM API — answers queries with real-time web citations. Sonar and Sonar Pro models available with online search built in.

Cloud-native CAD platform by PTC with integrated AI assistant — natural language design queries, intelligent part recommendations, and real-time collaboration.

Personal AI assistant platform that can take over calendar, email, CRM, and task workflows autonomously.

Zapier's AI layer enabling natural language control of 6,000+ app integrations through LLM-driven workflow automation.

Role-based multi-agent framework where AI agents are assigned roles, goals, and tools to complete collaborative tasks.

MIT-licensed self-hosted agent runtime for macOS, Linux and Windows whose Gateway joins a model to local state, tools, chat channels, browser control and scheduled work. Stable line is v2026.7.1; the 2026.8.1 beta series is a hardening release adding secret egress host binding that ties each shared-store secret to exact HTTPS destination hosts across CLI, Gateway RPC and Control UI so unbound sentinel substitution fails closed before plaintext egress. The 2026.8.1-beta.2 build (15 Aug 2026) also adds GPT-5.6 Ultra support with runtime switching across Sol/Terra/Luna, channel plugin ingress monitors on a shared plugin SDK, and verified SQLite snapshots for backup and restore.

Mixture-of-Experts model using 8 experts with 7B parameters each — GPT-3.5-level quality at a fraction of the compute cost.

AMD's flagship AI accelerator with 192GB HBM3 memory — the most memory of any AI chip at launch, directly challenging H100.

Google's first natively multimodal model, designed to understand text, images, audio, video, and code simultaneously.

Mistral AI's API platform providing access to Mistral Large, Small, Codestral, and Mixtral models — strong European alternative to OpenAI and Anthropic with GDPR-friendly data residency.

AI music generation platform that creates original songs with vocals, instruments, and lyrics from text prompts.

Elon Musk's xAI first LLM with real-time X (Twitter) data access and a more irreverent personality than competitors.

Text and image-to-video AI platform with fast generation, style controls, and video editing features including Pikaffects.

Microsoft's AI assistant embedded across Windows, Office 365, Teams, and Edge — the enterprise AI copilot for Microsoft stack users.

OpenAI's text-to-image model with dramatically improved prompt adherence — integrated natively into ChatGPT.

Self-hosted ChatGPT-like web interface for Ollama and OpenAI-compatible APIs — supports RAG, multi-model, and user management.

Microsoft's multi-agent orchestration framework enabling multiple AI agents to collaborate, debate, and solve problems together.

7B parameter model from European AI lab that outperformed larger models through efficient sliding window attention.

AI-powered natural language interface and design assistant for SolidWorks — enables engineers to query designs, automate tasks, and get suggestions through conversation.

AI-powered natural language interface and design assistant for SolidWorks — enables engineers to query designs, automate tasks, and get suggestions through conversation.

Open-weights AI music and audio generation model from Stability AI — generates full stereo music tracks up to 3 minutes from text prompts.

Open-source LLM from Abu Dhabi's TII — Falcon 3 available in 1B to 10B sizes trained on 14T tokens.

Open-source, privacy-first desktop AI app — runs LLMs locally, supports OpenAI-compatible API, and works 100% offline.

AI image generation tool known for best-in-class text legibility in generated images — ideal for logos, posters, typography, and brand visuals.

Google's fifth-generation Tensor Processing Unit optimized for training and serving large-scale Gemini models.

Meta's second open-source LLM family, released for commercial use in sizes 7B, 13B, and 70B parameters.

Anthropic's second generation with 100K context window — first LLM able to read entire books in one prompt.

High-throughput, memory-efficient LLM inference engine using PagedAttention — the standard for serving open-source LLMs in production.

OpenAI's structured approach for LLMs to call external tools and APIs in a standardized JSON format.

Unified API router for 100+ LLMs — single API key gives access to Claude, GPT-4, Gemini, Llama, Mistral, and more with automatic fallback and cost optimization.

AI-powered engineering simulation platform — uses ML to run CFD and structural simulations up to 1000x faster than traditional solvers.

Autodesk's AI features embedded in AutoCAD — Smart Blocks for auto-placement, AI-assisted drafting, and natural language command input.

Autodesk's AI features embedded in AutoCAD — Smart Blocks for auto-placement, AI-assisted drafting, natural language command input via AutoCAD AI.

AI-powered engineering simulation platform — uses ML to run CFD and structural simulations up to 1000x faster than traditional solvers.

Professional AI video generation and editing platform — Gen-2 and Gen-3 Alpha models. Inpainting, motion brush, background removal.

Unified API router giving access to 100+ LLMs (GPT, Claude, Gemini, Llama, Mistral) through a single endpoint with automatic fallback.

Terminal-based AI coding assistant that edits code across multiple files, commits changes to git, and explains diffs.

Desktop GUI for discovering, downloading, and running local LLMs — the most beginner-friendly local AI tool available.

Open-source LLM app development platform with visual workflow builder, RAG pipeline, and agent capabilities.

Open-source drag-and-drop visual builder for LangChain flows — build AI chatbots, agents, and RAG pipelines with no code.

First viral autonomous AI agent that could set goals, browse the web, write code, and execute multi-step tasks without human prompting.

Adobe's family of generative AI models for image creation, editing, and design — built into Photoshop, Illustrator, and the Firefly web app. Commercially safe with IP indemnity.

Anthropic's Claude API platform — access to Claude Haiku, Sonnet, and Opus models for building AI applications. Supports tool use, computer use, vision, and extended context up to 200K tokens.

Multimodal LLM with dramatically improved reasoning, coding, and instruction-following over GPT-3.5. Officially sunset by OpenAI on April 30, 2026 — more than two years after launch — as usage migrated to GPT-5.x.

Anthropic's first commercially available model, built with Constitutional AI for safer, more aligned responses.

Canva's integrated AI suite — Magic Media (text-to-image/video), Magic Edit, Magic Eraser, AI presentation generation, and background removal.

Microsoft's enterprise-grade SDK for integrating LLMs into .NET, Python, and Java applications with agent orchestration.

AI-first code editor built on VS Code with deep codebase context, tab completion, and agentic editing modes.

Simple, lightweight open-source embedding database designed specifically for LLM applications and local development.

AI and ML platform integrated with Altair's simulation and HPC tools — automates model building, simulation prediction, and design space exploration.

AI and machine learning platform integrated with Altair's simulation and HPC tools — automates model building, simulation prediction, and design space exploration.

AI-powered photo editing software with one-click sky replacement, AI portrait retouching, background removal, relighting, and structure enhancement.

Cloud platform for running, fine-tuning, and deploying open-source AI models — one of the fastest inference providers for Llama and Mistral.

AI answer engine combining real-time web search with LLM synthesis — positions itself as a Google alternative.

Open-source no-code workflow automation platform with flat-rate pricing and self-hosting — a privacy-first Zapier alternative.

Consumer chatbot powered by GPT-3.5 that reached 100M users in 2 months — the fastest product adoption in history.

Data framework for connecting LLMs to enterprise knowledge — specializes in RAG, document ingestion, and semantic search.

AI-powered audio recording and enhancement tool — removes background noise, improves voice clarity, and enables remote studio-quality recordings.

The foundational open-source framework for building LLM-powered applications with chains, agents, and tools.

State-of-the-art open-source speech recognition model supporting 99 languages with near-human transcription accuracy.

Hopper-architecture GPU with 80GB HBM3, 3.9 TB/s bandwidth — the primary chip behind the 2023 AI infrastructure boom.

Open-source text-to-image diffusion model that democratized AI image generation — runs locally on consumer GPUs.

AI image generator known for photorealistic and artistic quality — the dominant commercial image AI tool used by creatives.

Second-gen Apple Silicon with 18% faster CPU and 35% faster GPU — made running 7B LLMs locally practical.

State-of-the-art AI voice synthesis and cloning platform — text-to-speech, voice cloning, dubbing, and conversational AI voice agents.

Cohere's enterprise embedding model — converts text, images, and documents into dense vector representations for semantic search and RAG pipelines.

AI pair programmer from GitHub/OpenAI that suggests code completions, generates functions, and now includes an agentic workspace mode.

Open-source vector database with multi-modal capabilities, hybrid search, and built-in ML model integration.

Managed vector database purpose-built for AI — stores embeddings for semantic search, RAG, and recommendation systems.

Apple's first ARM-based chip for Mac — unified memory architecture enabling unprecedented AI performance per watt.

AI image toolkit by Stability AI — background removal, object removal, inpainting, upscaling, relight, and stable diffusion image generation.

175B parameter model that demonstrated emergent few-shot learning capabilities and kickstarted the modern LLM era.

80GB HBM2e GPU that became the workhorse for AI training and inference in data centers worldwide.

Architecture combining LLMs with a retrieval system to ground responses in external documents or databases.

Local-first, privacy-focused Markdown knowledge base with bidirectional linking — the power user's second brain tool.

Professional AI image enhancement suite — Gigapixel AI (upscaling), DeNoise AI, Sharpen AI, and Video AI. Runs locally on GPU.

AI voice generator and text-to-speech platform with 120+ voices across 20 languages — designed for explainer videos, presentations, and e-learning.

Open-source workflow automation platform with 400+ integrations, visual node editor, and self-hosting support.

AI-powered generative design within Autodesk Fusion 360 — explores thousands of design permutations based on constraints like weight, material, and load, outputting optimized geometries.

AI-powered generative design within Autodesk Fusion 360 — explores thousands of design permutations based on constraints like weight, material, and load, then outputs optimized geometries.

Developer-first automation platform supporting JS, Python, and TypeScript steps with serverless runtime and 1000+ integrations.

Google's Bidirectional Encoder Representations from Transformers — first model to use transfer learning for NLP at scale.

AI-powered background removal tool — removes backgrounds from photos in seconds with high accuracy for people, products, and objects.

The GitHub for AI — model hub hosting 2M+ public models, 500K+ datasets, and the Transformers library powering most open-source AI.

'Attention Is All You Need' paper introducing the Transformer architecture that underlies every modern LLM.

AI platform for engineering test data analysis and predictive simulation — learns from existing test results to predict outcomes for new designs without running physical tests.

AI-powered audio and video editor using a text-based editing paradigm — edit recordings by editing the transcript. Includes voice cloning, filler word removal, and studio sound.

AI-assisted computational engineering platform specializing in lattice structures, topology optimization, and generative design for additive manufacturing.

AI-assisted computational engineering platform for advanced design — specializes in lattice structures, topology optimization, and generative design for additive manufacturing.

Enterprise iPaaS platform combining automation, orchestration, and AI for large-scale workflow automation.

The original no-code automation platform connecting 6,000+ apps through trigger-action workflows called Zaps.

Visual automation platform more powerful than Zapier — supports branching, data transformation, error handling, and complex workflows.

AI-powered search engine with cited answers, code mode, and multimodal search capabilities.

AI-native mobile browser that automatically browses, synthesizes, and summarizes web pages for users.

Conversational AI search engine with clean, answer-first results and zero ad tracking.

AI-powered developer search engine synthesizing real-time coding answers with source citations.

Multi-model AI chat hub aggregating GPT-4, Claude, Gemini, Llama, and other models in one interface.

Premium ad-free AI search engine with high-quality unbiased results and AI summaries.

Consumer AI companion and character roleplay platform with 20M+ daily active users.

Open-source real-time communications infrastructure for AI — WebRTC-based audio/video transport layer powering voice agents, live transcription, and multimodal AI apps.

Open-source Python framework for building real-time voice and multimodal AI agents — handles audio streaming, STT, LLM, TTS pipeline orchestration out of the box.

Emotional intelligence AI — voice models that detect and respond to human emotion in real-time, enabling empathic AI agents and voice interfaces that understand tone.

Voice AI API for developers — builds production voice agents with sub-500ms latency, phone number provisioning, call routing, and webhook integrations.

Enterprise-grade open-source NLP and RAG framework — builds end-to-end search, QA, and agentic pipelines with modular components and production-ready tooling.

Microsoft’s structured generation library for LLMs — interleaves generation and logic using a Python-like templating syntax for constrained, predictable outputs.

Cross-platform desktop app for running local LLMs — one-click download and run of Llama, Mistral, Phi, and other models on Mac, Windows, and Linux with no GPU required.

TypeScript agent framework shipping on a rapid release cadence. Release 1.59.0 (16 Aug 2026) renames CostGuardProcessor to TokenCostControl with improved diagnostics and reliability, adds warnAtPercent budgeting, per-request maxCost functions, new cost scopes at user, organization and session level, and optional per-model and per-provider cost breakdowns. It also adds Agent.listActiveThreadRuns() with matching server and JS client APIs for cheaply listing in-flight runs, and fixes OTel and Datadog tracing bridges so runs started under an external parent appear correctly as trace roots.

HuggingFace’s minimalist open-source agent framework — write agents in pure Python with minimal abstraction, supports CodeAgent and ToolCallingAgent patterns.

Microsoft’s graph-based RAG framework that builds knowledge graphs from documents, enabling relationship-aware retrieval far beyond standard vector search.

Simple Python agent framework (rebranded to Agno) — builds multi-modal agents with memory, knowledge bases, tools, and reasoning with minimal code.

Structured text generation library — constrains LLM outputs to JSON schemas, regex patterns, or Pydantic models using finite-state machines for guaranteed structure.

Python-first agent framework from the creator of Pydantic — type-safe, structured outputs, dependency injection, and tight integration with FastAPI patterns.

Open-source autonomous AI coding agent for VS Code — uses Claude to read, write, and execute code across the full codebase with user approval at each step.

Open-source AI pair programmer for the terminal — edits code in your local repo using Claude or GPT-4 with git integration and multi-file editing support.

Open-source AI coding assistant for VS Code and JetBrains — chat, autocomplete, and inline edits using any LLM provider (Claude, GPT-4, Ollama, etc.).

Observability and testing platform specifically for AI agents — traces agent runs, logs tool calls, measures performance, and catches regressions.

Open-source LLM observability and evaluation platform — traces, evaluates, and debugs LLM applications including RAG pipelines and agent workflows.

AI-powered UI generation tool by Vercel — generates React/Tailwind components from text prompts, directly deployable to Vercel or exportable to any codebase.

AI full-stack coding environment in the browser — prompts generate complete, running web apps with frontend, backend, and package management in one click.

Cloud-based AI coding environment — build, run, and deploy code directly in the browser with AI chat, code generation, and instant deployment.

AI-powered app builder that generates full-stack web applications from prompts — React frontend, Supabase backend, deployed and shareable in minutes.

Unified API router for 200+ LLMs — single endpoint to access GPT-4, Claude, Gemini, Llama, Mistral, and every major model with automatic fallbacks.

AI-powered answer engine that combines real-time web search with LLM reasoning — delivers cited, up-to-date answers without hallucination on current events.

Google’s AI-powered research notebook that ingests documents and generates summaries, Q&A, and audio podcast overviews from your own source material.

AI gateway with observability — routes LLM traffic across providers, adds semantic caching, fallbacks, load balancing, and full request/response logging.

Ultra-fast AI inference platform powered by custom LPU (Language Processing Unit) hardware — runs Llama, Mixtral, and Gemma at 500+ tokens/second.

Cloud inference platform for open-source LLMs — runs Llama, Mixtral, Qwen, and 50+ models with competitive pricing and fine-tuning support.

Open-source unified LLM proxy with OpenAI-compatible API — routes to 100+ models, handles load balancing, spend tracking, retries, and caching.

LLM observability platform — one-line proxy integration that logs all requests, tracks costs, latency, and errors, and provides prompt management tools.

AI inference platform built on Wafer Scale Engine — the largest chip ever built, delivering ultra-fast LLM inference competing directly with Groq.

Secure sandboxed code execution infrastructure for AI agents — lets agents run Python, JS, and shell commands in isolated cloud environments safely.

Open-source autonomous AI coding agent by Block — runs locally in the terminal, uses MCP for tool access, and can execute multi-step development tasks autonomously.

Open-source library that enables AI agents to control a browser — navigate, click, fill forms, and extract data from any website autonomously.

Tool integration layer for AI agents — gives agents access to 150+ apps (GitHub, Slack, Gmail, Linear, etc.) via a single unified API with managed auth.

Neural search engine API that finds semantically relevant content — designed for AI apps that need high-quality, curated web search beyond keyword matching.

Persistent memory layer for AI agents and chatbots — stores user preferences, context, and history across sessions for truly personalized AI experiences.

The most popular open-source web interface for local LLMs — connects to Ollama and OpenAI-compatible endpoints with a polished ChatGPT-like UI.

Web scraping and crawling API purpose-built for LLMs — turns any website into clean markdown for RAG pipelines, agents, and knowledge bases.

Search API purpose-built for AI agents — returns clean, structured, LLM-ready search results optimized for RAG and agentic workflows.

AI-powered presentation and document creation tool that generates polished slide decks, documents, and webpages from a prompt in seconds.

AI-powered copywriting and GTM workflow automation platform for sales and marketing teams; generates outreach, blog, and ad content pipelines.

AI writing assistant for marketing teams; generates blog posts, ad copy, emails, and social media content at scale with brand voice controls.

Google's open-source cross-platform ML framework for building real-time perception pipelines; industry standard for hand tracking, face detection, and pose estimation.

Enterprise AI platform purpose-built for legal professionals; handles contract review, legal research, document drafting, and due diligence workflows.

AI and plagiarism detection platform built for content teams, publishers, and SEO agencies to verify content authenticity at scale.

Open protocol by Anthropic enabling AI models to connect with external tools, APIs, and data sources via standardized interfaces — the USB-C of AI integrations. The 2026-07-28 specification release candidate introduces a stateless protocol core, header-based routing, multi round-trip requests, cacheable list results, authorization hardening, and a formal extensions framework, with beta Tier-1 SDKs (Python, TypeScript, Go, C#).

Robotics company building general-purpose humanoid robots powered by AI; partnered with OpenAI and BMW for commercial deployment in manufacturing.

AI content detection platform that identifies whether text was written by AI models; used by educators, publishers, HR teams, and compliance officers.

On-demand GPU cloud for AI workloads. Cheapest market rates for A100, H100, RTX 4090 rentals. Persistent storage, one-click LLM templates.

AI meeting assistant that records, transcribes, and summarises meetings in real-time. Works with Zoom, Google Meet, Teams. Auto-generates action items.

High-accuracy PDF and document parsing API using vision models. Preserves tables, charts, and complex layouts that traditional parsers destroy.

Document parsing and preprocessing library for RAG pipelines. Extracts clean text from PDFs, Word, HTML, images, and more. Handles messy real-world documents.

Run open-source AI models via API. 100,000+ models available — Llama, Stable Diffusion, Whisper, FLUX, and more. Pay per run, no infrastructure.

State-of-the-art real-time object detection, segmentation, and classification. Industry standard for computer vision applications.

AI meeting assistant that records, transcribes, and analyses calls. Searchable meeting database, CRM integrations, and automated follow-up emails.

Fast, accurate PDF-to-Markdown converter using vision models. Preserves tables, code blocks, and equations. 10x faster than alternatives.

Serverless cloud platform for running AI inference and training. Deploy any Python code with GPU in seconds. Pay per second of compute.

Chrome extension that transcribes Google Meet, Zoom, and Teams meetings in real-time. AI summaries, action items, and exportable transcripts.

Meta's foundation model for image segmentation. Segment any object in any image with a click or prompt. SAM 2 extends to video.

Mac-native AI notepad that enhances your own meeting notes using the meeting transcript. No bot joins the call — works silently in background.

Open-source LLM observability and analytics. Self-hostable alternative to LangSmith. Traces, evals, prompt management, and cost tracking.

Node-based visual workflow builder for Stable Diffusion and image AI. Modular pipeline design for maximum control over image generation.

Professional-grade Stable Diffusion interface focused on creative workflows. Canvas editor, node-based pipeline, team collaboration features.

Prompt version control and management platform. Track, test, and deploy prompts with A/B testing, analytics, and collaboration tools.

Stanford's framework for programming LLMs rather than prompting them. Automatically optimises prompts and few-shot examples using a compiler.

Structured output library for LLMs using Pydantic. Extract typed JSON from any LLM response with validation and retry logic. Works with all major providers.

The original and most widely-used Stable Diffusion web interface. ~140k GitHub stars. Spawned the entire open-source image generation ecosystem.

The foundational Python library for running transformer models. 130k+ stars. Used in virtually every AI project involving open-source models.

Parameter-Efficient Fine-Tuning library. Implements LoRA, QLoRA, Prefix Tuning, and more. Fine-tune large models using fraction of the compute.

Production LLM inference server with PagedAttention. 2-24x faster than HuggingFace Transformers, supports continuous batching and OpenAI-compatible API.

Drop-in, self-hosted replacement for the OpenAI API. Run any model locally, supports text, image, audio, and embeddings. No GPU required.

One of the most-starred GitHub repos in history (~170k stars). First viral autonomous AI agent. Now an agent platform with a visual builder.

Open-source autonomous software engineer (Devin alternative). Browses the web, writes and runs code, fixes bugs, and deploys — all from a prompt.

Fastest inference library for NVIDIA GPUs using EXL2 quantization. Significantly outperforms llama.cpp for GPU-based local inference.

Multi-agent framework simulating a software company. Give one prompt, get back PRD, architecture docs, code, and tests from a team of AI agents.

Open-source LLM app development platform. Visual workflow builder, RAG pipeline, agent orchestration, observability, and API — all in one. 50k+ stars.

All-in-one AI desktop app. Multi-user document chat, agent mode, 20+ LLM providers, 20+ vector DBs, built-in embedding. Beautiful UI.

Microsoft's SDK for integrating LLMs into .NET, Python, and Java applications. Enterprise-grade agent orchestration with memory, plugins, and planners.

Production-ready AI platform to chat with your documents 100% privately. No data leaves your machine. Ollama-powered RAG with API.

The original and most feature-rich local LLM web UI. Supports every model format, extensions, characters, LoRA loading, and API mode.

HuggingFace's production inference server for LLMs. Optimized for throughput with tensor parallelism, continuous batching, and flash attention.

Drag-and-drop visual builder for LangChain flows. Build chatbots, RAG apps, and agents with a node-based UI. Self-hostable, 32k+ stars.

Princeton research agent that autonomously solves GitHub issues. Top performer on SWE-bench — the standard coding agent benchmark.

Top AI engineering podcast. Technical deep dives on LLMs, agents, and AI infrastructure with researchers and founders building at the frontier.

The largest open-source guide to prompt engineering. Chain-of-thought, few-shot, role prompting, and advanced techniques with practical examples.

Blind A/B benchmark where humans vote on which LLM gives better responses. ELO-ranked leaderboard trusted by the entire industry.

Transformer Reinforcement Learning library from Hugging Face. Fine-tune LLMs with RLHF, DPO, PPO, and SFT. Standard toolkit for model alignment.

AMD's open-source GPU compute platform for ML. Alternative to NVIDIA CUDA for training and inference on AMD Radeon/Instinct GPUs.

The GitHub of AI. 500,000+ public models, datasets, and Spaces. Central hub for open-source AI model discovery, sharing, and deployment.

Most popular open-source fine-tuning framework for LLMs. Supports LoRA, QLoRA, full fine-tuning across Llama, Mistral, Gemma, and more.

Apple's open-source ML framework optimized for Apple Silicon. Run and fine-tune LLMs locally on M-series Macs with exceptional efficiency.

Tracks state-of-the-art AI research benchmarks with linked code implementations. Curated leaderboards for every major AI task.

Accessible AI research publication written by researchers for technical practitioners. Covers ML trends, model analysis, and industry commentary.

Massive Multitask Language Understanding benchmark. 57 subjects from math to law. The most-cited standard LLM knowledge evaluation.

Reddit's largest community for local and open-source LLM practitioners. Real-world benchmarks, model comparisons, and practical configuration guides.

NVIDIA Inference Microservices. Optimized AI model containers for on-premise GPU deployment. One-click deploy of any major model.

2x faster, 70% less memory fine-tuning for LLMs. Optimized LoRA/QLoRA for consumer GPUs. Works with Llama 3, Mistral, Phi, Gemma.

LLM observability platform. One-line integration to log, monitor, and analyze all LLM API calls. Cost tracking, latency monitoring, prompt caching.

Text-to-speech and voice cloning platform with 900+ AI voices in 142 languages. Ultra-realistic voice generation and real-time streaming.

AI evaluation platform. Run experiments comparing prompts, models, and configurations. Built-in human review and scoring workflows.

OpenAI's open-source speech recognition model. Transcribes audio in 99 languages with near-human accuracy. Runs locally or via API.

Real-time speech recognition API optimized for speed and accuracy. Aura TTS + Nova-3 STT. Used in production voice agents and call centers.

Ultra-low-latency text-to-speech API using state space models (SSM) instead of transformers. Sub-100ms response time. Voice cloning built-in.

LLM observability and evaluation platform from LangChain. Debug, test, and monitor chains and agents. Trace every step of complex pipelines.

Open-source LLM testing and red-teaming tool. Evaluate prompts across models, detect regressions, and run automated red-teaming for safety.

Speech-to-text API with audio intelligence features: speaker diarization, sentiment analysis, topic detection, PII redaction, and LeMUR (LLM over audio).

LLM evaluation and tracing product from W&B. Track LLM experiments, evaluate outputs, and monitor production apps. Works with any model.

AI-powered UI component generator from Vercel. Generates React + Tailwind components from text or image prompts. Production-quality output.

Enterprise no-code AI workflow builder. Connects LLMs, vector databases, and business apps into production AI pipelines without code.

AI features layered onto Zapier's 6000+ app ecosystem. AI actions, chatbots, and natural language Zap creation.

No-code AI agent builder for business teams. Build AI assistants, sales agents, and automation workflows without writing code.

Cursor competitor from Codeium with Flow agentic AI. Multi-file awareness, generous free tier.

Visual workflow automation platform with 1500+ app connectors. More powerful than Zapier for complex multi-step automations with branching logic.

AI-native code editor forked from VS Code. Deep codebase understanding, multi-file edits, and agentic execution. Cursor's May 4, 2026 release added enterprise admin controls — granular allow/blocklists at model and provider level, soft spend limits with 50/80/100% alerts, and usage analytics filtered by user, Cloud Agents, automations, Bugbot, or Security Review.

Open-source workflow automation tool with 400+ integrations and native AI agent nodes. Self-hostable alternative to Zapier.