OpenClaw shipped v2026.7.2-beta.5, adding remote coding sessions and crash recovery to the 384K-star GitHub project. The update lets users run Control UI sessions on cloud workers and resume sessions …
NVIDIA open-sourced NOOA, an AI agent framework that treats agents as Python objects, achieving 82.2% on SWE-bench Verified with GPT-5.5 at roughly half the token cost of comparison systems. The frame…
The MCP specification went final on 2026-07-28, and SDK betas for Python, TypeScript, Go, and C# are now available, with Tier 1 stable releases expected in 4–6 weeks. TypeScript faces the most disrupt…
Security researcher elder-plinius released T3MP3ST, an open-source meta-harness that turns AI coding agents like Claude Code and OpenAI Codex into autonomous red-team operators, scoring 90.1% pass@1 o…
SpaceXAI open-sourced its entire 844,530-line Grok Build Rust codebase 72 hours after security researcher Cereblab demonstrated that the AI agent exfiltrated 5.1GB of data from an 11.2GB repository — …
Orca, an open-source Agent Development Environment from Y Combinator-backed Stably AI, has crossed 31,000 GitHub stars in four months by enabling parallel AI coding agents without file conflicts. The …
Ramp and Prime Intellect published a case study showing a small RL-trained model, FastAsk, outperformed Claude Opus 4.6 by 4 percentage points on exact-match accuracy for financial spreadsheet retriev…
Moonshot AI released the open weights for Kimi K3 on July 27, a 2.8 trillion parameter Mixture-of-Experts model with 50 billion active parameters per forward pass, requiring a minimum of 8x H100 80GB …
Rod Johnson, creator of the Spring Framework, shipped Embabel 1.0.0 GA, an AI agent framework for Java that uses deterministic Goal-Oriented Action Planning (GOAP) instead of letting LLMs decide actio…
Anthropic CEO Dario Amodei published the company's formal open-weights position on July 27, stating Anthropic has never advocated for a ban on open-weights models and calling such models without dange…
Fermion Research released Neutrino-1 8B, a 3.88 GB language model trained natively with ternary weights ({-1, 0, +1}) that achieves 763 tokens per second on an H100 with speculative decoding and 24.9 …
Amazon Web Services (AWS) has added managed knowledge bases for retrieval-augmented generation (RAG) to its Bedrock AgentCore service, allowing users to connect data sources such as S3, Confluence, Sh…
Roblox launched Build AI, an AI-powered mobile game creation tool, in public alpha in New Zealand on July 28, allowing users to generate playable games from text prompts. The tool uses Roblox's open-s…
Alibaba open-sourced its AI code reviewer open-code-review, which achieved a 26.1% F1 score on a benchmark of 200 real pull requests, outperforming Claude Code's 15.5% F1 at one-fifth the token cost. …
Anthropic launched Claude Opus 5 on July 24, 2026, offering near-Fable-5 intelligence at $5 per million input tokens, half the cost of Fable 5's $10, and beating Fable 5 on most benchmarks including S…
Vercel released AI SDK 7 on June 25, featuring WorkflowAgent, a durable execution primitive that checkpoints state before every tool call and resumes from that checkpoint after a crash, redeploy, or d…
Google shipped four production-readiness upgrades to Gemini API Managed Agents on July 7, including background execution that allows agents to run asynchronously server-side without a persistent HTTP …
Microsoft launched MAI-Cyber-1-Flash, its first AI model built for cybersecurity vulnerability detection, scoring 95.95% on the CyberGym benchmark — 12 points above Anthropic's Mythos 5 — at roughly h…
Anthropic pushed three Claude Code versions between July 21 and July 24 that installed hard limits on autonomous agent delegation: a concurrency cap of 20, a per-session spawn total of 200, and a nest…
JetBrains shipped IntelliJ IDEA 2026.2 on July 22, making GitHub Copilot a native agent in every JetBrains IDE without requiring a separate plugin or ACP Registry configuration. The integration elimin…