The AI Daily Brief: Artificial Intelligence News and Analysis Claude Sonnet 4.5 Can Code Autonomously for 30 Hours 🤯
817 snips
Sep 30, 2025 Anthropic's Claude Sonnet 4.5 is revolutionizing AI coding with its ability to autonomously code for up to 30 hours, far surpassing earlier benchmarks. Enhanced features like persistent memory and runtime constraints allow it to tackle complex tasks efficiently. Initial reactions show varied experiences with its coding quality. The model not only excels in coding but also demonstrates improvements in math and finance tasks. There are exciting implications for AI's potential to write future iterations of itself, marking a significant leap in autonomous AI capabilities.
AI Snips
Chapters
Transcript
Episode notes
Sonnet 4.5 Targets Coding Leap
- Anthropic's Sonnet 4.5 targets coding and agentic workflows with major benchmark gains.
- It claims improved reasoning, math, and computer use over prior Claude versions.
Benchmarks Show Notable Gains
- Benchmarks show Sonnet 4.5 ahead on multiple coding metrics versus competitors.
- Reported gains include SweetBench and TerminalBench improvements versus GPT-5 and Opus.
Teams Rebuilt Agents Around Sonnet
- Several agentic coding teams switched Sonnet 4.5 into production and rebuilt tooling around it.
- Cognition reported faster multi-hour sessions and context-aware behavior changes in their Devon agent.