Latest Anthropic News

Mods are small TypeScript functions that alter Claude Code behavior, such as rewriting prompts or blocking commands, and can be shared as plugins.

GLM-5.3 can autonomously build end-to-end cyber exploits and was released without meaningful safeguards to limit misuse, unlike other frontier models.

Anthropic says Claude Sonnet 5.5 improves on Claude Sonnet 5, runs over 30% faster, and costs up to 30% less for most tasks.

A new portal lets developers submit plugins to the Claude directory, package MCP connectors and Agent Skills, track review status, and view usage data once live.

Claude Tag can now use your connectors for requests you make in a channel, while others cannot, and you control how output is presented.

The latest Opus model is priced and trained to optimize costs for current developer coding practices.

Early results from a new life sciences research lab show Claude agents identified an enzyme system whose function remains unknown.

Claude Opus 5.5 leads in agentic coding and knowledge work and costs 40% less to run than Opus 5 on typical workloads.

Developers welcomed the change as relief from a compatibility problem.

Anthropic is an AI safety and research company that works to build reliable, interpretable, and steerable AI systems.