The model offers flexibility in thinking modes within a single architecture. This release occurred on February 24, 2025, alongside vision support added to Claude Haiku 3.5 for image analysis. Both updates enhance Claude's API capabilities.
The tool reasons like a human researcher, traces data flows, and assigns severity ratings after multi-stage verification. It emphasizes human-in-the-loop approval with confidence scores. Aimed at countering AI-assisted attacks, it launched February 21, 2026.
The model improves planning, sustains longer agentic tasks, handles larger codebases, and excels in code review and debugging. It integrates with Excel upgrades and a PowerPoint research preview for workplace tasks. Available now on claude.ai, API, and cloud platforms at unchanged pricing of $5/$25 per million tokens.
The tool reviews codebases for subtle security flaws at scale, showcasing the model's prowess beyond expert human review. It builds on recent Claude updates for coding and agentic tasks. This enhances AI-assisted software security practices.
These quality-of-life improvements reshape developer workflows by enabling better pre-coding planning, persistent memory for error correction, and frictionless voice prompting. The updates build on weekly feature shipments, including remote sessions and agent orchestration. Adoption has surged with major model upgrades since early 2026. The video overview was published on March 12, 2026.
The constitution emphasizes four core principles: being broadly safe by not undermining human oversight, broadly ethical through honesty and good values, compliant with Anthropic's guidelines, and genuinely helpful to users and operators. It addresses Claude's psychological security and sense of self amid uncertainties in AI development. The full text is now public, with plans for additional training and transparency materials.
The updates, announced March 11, 2026, allow actions in one application to be informed by the other, improving workflow integration. Users of Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry can now connect through an LLM gateway. These changes expand Claude's utility in productivity software for enterprise environments.[2]
The investment, revealed on March 12, 2026, supports partners building tools and applications powered by Claude models. This move follows recent product enhancements and aims to accelerate adoption across industries. It builds on Anthropic's strategy to create a robust partner ecosystem amid competitive AI developments.[2]
The update reduces bundle size by 510 KB, hides CLAUDE.md comments from Claude during auto-injection, fixes slow exits on background tasks, and improves rename functionality during processing. Prompt input re-renders drop by 74%, startup memory by 426 KB, and remote control polling cuts server load 300x. Security tightens and Claude Code experiences improve for more reliable operations across platforms. These changes target enterprise and developer workflows.
Thoughtworks hosts live AI news rundown on March 12, 2026
Weekly live session by Thoughtworks engineers Ben O'Mahony and Danilo Sato covers the most important AI developments, filtered for practical building insights. Published yesterday.
Ai2 releases Olmo Hybrid, 7B open model with 2x data efficiency
Ai2 launched Olmo Hybrid, a 7B-parameter open model family blending transformer attention and linear recurrent layers. It matches Olmo 3's MMLU accuracy using 49% fewer tokens, per
Washington State Passes Three Key AI Bills on Legislature Adjournment Eve
HB 1170 mandates AI disclosure, HB 2225 requires chatbot safety for kids including self-harm protocols, and SB 5395 regulates AI in health insurance. Bills received final...
Microsoft integrates Anthropic's CoWork with Claude Sonnet into Copilot
Microsoft is adding Anthropic's CoWork feature, powered by Claude Sonnet models, to Copilot in Microsoft 365 for the first time, enhancing enterprise AI collaboration tools.