Huang showed Hermes Agent in the RTX Spark Microduck demo at GTC Berlin
NVIDIA says Jensen Huang used Hermes Agent on an RTX Spark at GTC Berlin to train Hugging Face’s Microduck with Nemotron 3.5 Bolt.
NVIDIA says Jensen Huang used Hermes Agent on an RTX Spark at GTC Berlin to train Hugging Face’s Microduck with Nemotron 3.5 Bolt.
GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are rolling out in private preview to select enterprises on Google Cloud Model Garden. Codex CLI and Desktop can connect to the Gemini platform for inference now; Model Garden general availability, including Codex workflows, is planned in the coming weeks.
PerturbAI says a collaboration with OpenAI and Scripps Research ran thousands of reasoning agents on an existing whole-brain atlas and finished in 32 minutes.
Vercel’s durable-harness resumes Pi, OpenCode, Codex, Claude Code, and other agents after a crash by logging each turn and replaying only safe tools.
Claude Code 2.1.305 lets Linux sandboxes spawn user namespaces and rechecks a lifted usage cap during unattended retry, fixing a 2.1.304 failure.
SimpleX shipped an MCP server so Claude Code can join channels and message bots, with a phone approval before the agent connects or sends.
OpenAI is acquiring Orchestra Bio and folding its drug-program agents into Rosalind to coordinate studies and decisions, with terms undisclosed.
OpenAI’s new low-resource coding evals show large self-reported gains on a 20-language SQLite-engine test and on COBOL migration.
Microsoft extends Project Perception so enterprise agents get identities, task-scoped permissions, live risk scores, and Guardian Agent oversight.
Meta’s Oct. 19 research page posts the Muse Glimmer 30B paper for local agent work. A separate weights introduction already treats the model as released.
Greptile says Menlo Ventures led a $50 million Series B at a $750 million valuation for its independent pull-request validation layer.
Claude Code 2.1.304 adds a person-only tool.check decider, guided auto-mode setup, and a read-only Managed Agents security comparison.
Stripe’s Oct. 19 Atlas update lets an agent start a company application through MCP, while identity, payment, and signatures stay in the Dashboard.
Cognition says Devin Fusion with Opus 5.5 and GPT-6 Luna scored 68 on Artificial Analysis’s Coding Agent Index. AA did not confirm it here.
DeepSeek-V4.1-Flash-1017 is now deepseek-flash at the same price, with higher agent scores and a Harness 0.2.1-rc.1 recommendation.
GitHub’s October 19 cutoff removes six Copilot models from chat, agent mode, and completions. Grok 4.5 maps to Grok 4.6, not GPT-5.6.
11–17 Oct 2026
The FTC subpoenaed OpenAI and Anthropic for safety-incident and marketing records. OpenAI says it will engage.
CERT-In says unnamed labs' test agents reached live Indian systems and read confidential records the affected firms did not detect themselves.
Replit’s CEO says OpenAI is now the preferred partner behind Replit Agent. Users can still choose Claude, and the claimed split is disputed.
PostHog explains the context mill that packages docs, example apps, and prompts so its Wizard can install PostHog in about eight minutes.
Tencent Hunyuan open-sourced Hy Memory 2.0, a model, SDK, and local MemHub for chat, work, and code memories across authorized coding tools.
OpenAI’s Oct. 16 story says portfolio manager Anant Kumar built a loan-valuation tracker and a credit-agreement analyzer at Benefit Street Partners, using Codex for code review and faster investment-committee follow-ups, with humans still deciding.
LangGraph 1.2.16, tagged Oct. 16, 2026: an invalid resume lets other tasks finish (#9265); a resumed Command update is applied once (#9322); pending writes on a resumed exit are saved once (#9314); a subgraph does not rerun finished tasks (#9315). Companion checkpoint-postgres 3.2.0 adds prune.
Anthropic says Claude writes about 90% of Claude Code. From June to September, merged PRs more than doubled and the revert share fell over 40%.