opus 4.5 and gpt-5 agent achieves 72.6% success, roughly human-level
on the OSWorld benchmark that tests whether AI can complete real computer tasks across various operating systems and everyday apps
this feels like a real milestone.
i believe we're nearing the point where computer-using agents can handle most desktop office work
We're proud to partner with @ENERGY and the Trump Administration on the Genesis Mission.
By combining DOE's unmatched scientific assets with our frontier AI capabilities, we'll support American energy dominance as well as advance and accelerate scientific productivity.
Anthropic released Claude Opus 4.5 (claude-opus-4-5-20251101) as their smartest model at $5/$25 per million tokens with top performance for coding, agents, and computer use, launched new beta features for developers, and expanded Claude for Chrome and Claude for Excel to more users
- Claude Opus 4.5 scored higher than any human candidate ever on Anthropic's difficult engineering exam in the 2-hour time limit, performs best across 7 out of 8 programming languages on SWE-bench Multilingual, works better at tasks like research and handling slides and spreadsheets, and has the strongest resistance to prompt injection attacks among frontier models
- Programmatic tool calling lets Claude invoke tools from code execution reducing token usage by 37%, tool search tool lets Claude load tools on-demand cutting token usage by 85% while improving Opus 4 accuracy from 49% to 74%, tool use examples provide sample tool calls in tool definitions to improve accuracy from 72% to 90% on complex parameter handling, and effort parameter lets developers pick between speed or thoroughness with medium effort matching Sonnet 4.5 while using 76% fewer tokens
- Claude for Chrome moved from research preview to beta for all Max users with attack success dropping to about 1%, Claude for Excel is now in beta for Max/Team/Enterprise users with pivot tables, charts and file uploads
- Claude Code gained Plan Mode improvements that build more precise plans and ask clarifying questions upfront, and Claude Code is now available in the desktop app letting you run multiple local and remote sessions in parallel
- For paid plan users with code execution enabled, Claude now automatically manages long conversations by summarizing earlier messages when approaching context window limits while preserving full chat history, and Max and Team Premium users now have roughly the same number of Opus tokens as they previously had with Sonnet with Opus-specific caps removed
🚨 Clause Opus 4.5 just dropped!
#1 on SWE-Bench Verified → 80.9%
(beating GPT-5.1 and Gemini-3-Pro)
- Crushes multi-system bugs with reasoning
- Creative fixes that benchmarks didn’t even expect 💀
- Huge jump in coding + agents + long-running tasks
Anthropic just dropped a monster