Anthropic Launches Claude Sonnet 4.6

#1 𝕏

Claude launched Sonnet 4.6, its most capable Sonnet model yet with upgrades across coding, long-context reasoning, agent planning, knowledge work and design. It also features a beta 1M-token context window.

Also covered by: @v0, @Aravind Srinivas, @Boris Cherny, @Mike Krieger

#2 📝 Simon Willison

Introducing Claude Sonnet 4.6 - Anthropic released Sonnet 4.6, claiming performance comparable to Opus 4.5 while keeping lower Sonnet pricing; Simon updated llm-anthropic to support the new models and ran pelican SVG generation tests showing Sonnet often adds a top hat.

Also covered by: @v0, @Aravind Srinivas, @Boris Cherny, @Mike Krieger

#3 𝕏

Qwen launched the Qwen3.5-397B-A17B model on @OpenRouterAI, inviting users to try out the 397B-parameter release and thanking the community for their support.

#4 📝 Simon Willison

Qwen3.5: Towards Native Multimodal Agents - Alibaba released Qwen 3.5 series including a large open-weight Mixture-of-Experts model (Qwen3.5-397B-A17B) that activates 17B parameters per forward pass and a proprietary hosted Qwen3.5 Plus with 1M token context; Simon tested pelican image outputs from both hosted and open models.

#5 𝕏

Google Research open-sourced a dataset of 2 million question/answer pairs to accelerate development of next-generation intuitive navigation and robotics.

#6 𝕏

DeepLearning.AI SpaceX acquired xAI (maker of the Grok models), creating a $1.25 trillion private company. They plan to integrate AI into space operations and eventually build solar-powered data centers in orbit.

#7 𝕏

Anthropic signed a first-of-its-kind MOU with the Government of Rwanda to deploy its AI tools across health, education, and other public sectors.

#8 𝕏

Harrison Chase launched agent-debugger, a Textual UI terminal debugger for LangGraph/LangChain agents that unifies agent-level visibility (state, messages, tool calls, store snapshots, semantic breakpoints) with Python-level debugging (line breakpoints, stepping, stack, local...

#9 𝕏

LlamaIndex 🦙 launched page-level extraction in LlamaExtract, mapping data to specific pages with bounding boxes and audit-ready citations, turning 200-page docs into skimmable, structured insights.

#10 𝕏

Cursor launched AWS agent plugins, giving its AI assistant built-in skills and tools to architect, deploy, and operate applications on AWS.

#11 𝕏

Cursor demos how to optimize React apps using the Vercel plugin in a concise video, sharing key best practices for performance and build efficiency.

#12 𝕏

Philipp Schmid picks ACP as his dark-horse contender to explode next, noting you can now effortlessly spin up an ACP server for any deep agent via Zed’s Agent Client Protocol & LangChain integration.

#13 𝕏

Harrison Chase warns that understanding real‐world usage of your production agent is tough but crucial. LangSmith Insights (from LangChain) solves this by surfacing detailed user interaction analytics so you can continually improve the agent experience.

#14 𝕏

Guillermo Rauch partnered with Socket Security, Snyk, and GenDigital to continuously audit skills.sh for security vulnerabilities, covering over 62,000 open-ecosystem skills.

#15 ▶️

Did My Claude Code AI Agent Automate a Six Figure Job?

All About AI

#16 𝕏

Google Research built MapTrace, a fully automated generative AI pipeline that produces 2 million high-quality map-path pairs to close the spatial grammar data gap. Fine-tuning Gemini 2.

#17 𝕏

NVIDIA AI and @basetenco built an optimized inference stack on NVIDIA Blackwell (NVFP4, TensorRT LLM & Dynamo) running gpt_oss_120b to return 30M+ physician minutes. It delivers 10× cost savings and 65% faster clinical note generation.

#18 𝕏

Logan Kilpatrick added billing limits to Google AI Studio’s Gemini API to curb account abuse and will work with the team to make the process more seamless for users.

#19 𝕏

claire vo 🖤 breaks down GPT-5 3 Codex vs Claude Opus 4.6 in her latest video and blog post, comparing their code-generation benchmarks, feature sets, and real-world API use cases.

#20 𝕏

Claude is hosting a one-year birthday bash for Claude Code on Feb 21 in San Francisco, featuring live demos, top hackathon projects, and cake. Spots are limited—builders can RSVP now to showcase their own work.

#21 𝕏

DeepLearning.AI Andrew Ng urges Hollywood and AI developers to collaborate on shared guardrails around generative AI, based on conversations at Sundance. The Batch also highlights SpaceX’s acquisition of xAI for orbital AI data centers, Claude Opus 4.

#22 𝕏

Philipp Schmid released an open-source LangChain demo that “closes the loop” on autonomous agents by pairing a GPT-4 doer with a GPT-4 critic—logging actions, scoring them against objectives, and feeding back refinements via memory and metrics for self-aware iterative improve...

#23 𝕏

Qwen launched Qwen3.5-Plus on Poe, offering faster inference, improved instruction-following and an expanded context window for more robust responses.

#24 𝕏

Santiago lays out a guardrail framework for safe agentic coding—spinning up agents in sandboxed environments, integrating them into CI/CD pipelines with automated tests and code-review bots, using feature flags and audit logs, and keeping human-in-the-loop oversight at every ...

#25 𝕏

Boris Cherny announced a new extended_context_tokens feature in Claude’s model config, unlocking a 1 million-token context window for ingesting massive docs (books, logs, datasets) in one go.