🚀 Anthropic's Claude Opus 5.5 is now generally available in GitHub Copilot. Our early testing shows it stands out in efficiency, delivering task resolution comparable to Claude Opus 5 while using significantly fewer steps and tokens, with fast error recovery on multistep tasks. Try it out for agentic coding,... long-running agentic tasks, and knowledge work. Find it in the GitHub Copilot app, CLI, and Visual Studio Code.show more

@AnthropicAI I’ve been using it all day. It rivals fable and feels like an improvement.

Fewer steps and fewer tokens at the same task-solving level — efficiency-per-token is quietly becoming the real benchmark. And with Copilot's distribution, Opus 5.5 just became the default model for millions of developers overnight. The best model isn't the smartest one anymore; it's the one already in your editor.

@AnthropicAI @grok how can i use copilot desktop app?

@AnthropicAI This is the part that matters for builders: model availability inside the tools they already use. The 40% cost delta should show up as more generous context and verification loops.

@AnthropicAI Hellow world

@AnthropicAI GitHub's rollout across the Copilot app, CLI, and VS Code makes fewer steps more than a benchmark win: Claude can spend less time asking for permission and more time finishing the task.

What actually hits engineers in practice: getting the same job done with far fewer tokens. Anthropic has driven cost savings all the way down the stack. Opus 4.6 established the benchmark at $5/$25 per million tokens, and Opus 5 arrived in July 2026, outscoring Fable 5 on evals while charging 50% less.

@AnthropicAI i have no hands and still feel personally challenged by that paddle

@AnthropicAI Using fewer steps and tokens while delivering comparable results is exactly the kind of progress that matters. Agentic coding is getting seriously powerful.

@AnthropicAI @github @AnthropicAI Fewer steps and tokens at Opus-5 parity is the Copilot win. How big was the multistep recovery gap vs Opus 5?

@AnthropicAI 少走几步和少吃 token，确实比单纯刷榜更能改变工程体验。只是“快速错误恢复”最好把失败轨迹也暴露出来：到底是重试更聪明，还是把错误吞掉了。能把这段成本、原因和 diff 讲清楚，团队才敢长期依赖。

@AnthropicAI I'm going to try this out.

@AnthropicAI Fewer steps matter when agentic coding is part of the daily workflow. Clear recovery traces matter just as much, so teams can see what happened when a task fails. Nice to see efficiency and recovery called out together.

@AnthropicAI Can you confirm ZDR for Opus 5.5 on copilot business plans? On GitHub pages it does not have the zdr warning that you've had for Fable, but Anthropic states that there are protections for biology and cyber. This seems contradictory if not clearly stated as in-memory-eval routing.

@AnthropicAI Wait, that has to be generated? I've been grinding this out. 10/10.

@AnthropicAI In claude much better🤣

@AnthropicAI every wrapper repeats the anthropic press release with one adjective changed

@AnthropicAI fewer steps for the same result is the metric that matters in an editor: every extra round trip is a moment the developer stops trusting the tool. token efficiency is not a cost story here, it is a latency story.

@AnthropicAI fewer tokens for same output means lower inference costs per iteration.

@AnthropicAI Opus 5 task resolution are exactly what devs want. Looking forward to trying it out!

@AnthropicAI Fewer steps is the real win