Claude Opus 4.1
Anthropic's upgraded Claude Opus 4 model for coding, agents and reasoning
Claude Opus 4.1 is a large language model from Anthropic, released as an upgrade to Claude Opus 4 focused on agentic tasks, real-world coding and reasoning. The company reports a 74.5% score on SWE-bench Verified and says it improves research and data analysis work, especially detail tracking and agentic search.
- Upgrade to Claude Opus 4 for agentic tasks, coding and reasoning
- Anthropic reports 74.5% on SWE-bench Verified
- Same pricing as Opus 4; API name claude-opus-4-1-20250805
- Available via Claude, Claude Code, Amazon Bedrock and Vertex AI
- Hybrid reasoning with extended thinking up to 64K tokens on some tests
What it is
Claude Opus 4.1 is a general-purpose language model made by Anthropic. It arrived as an upgrade to Claude Opus 4, aimed at agentic tasks, everyday coding and reasoning. The company reports it reaches 74.5% on SWE-bench Verified, a benchmark of real software issues, and says it also improves in-depth research and data analysis, particularly detail tracking and agentic search. Like the rest of the Claude 4 line, it is a hybrid reasoning model: it can spend extra computation on extended thinking before answering. Anthropic recommends it over Opus 4 for all uses and kept the same pricing.
How it came about
Anthropic released Opus 4.1 as a step up from Claude Opus 4, saying larger improvements were planned for the following weeks. It reached paid Claude users, Claude Code, the Anthropic API, Amazon Bedrock and Google Cloud's Vertex AI at launch. The company pointed developers to the model identifier claude-opus-4-1-20250805 and published a system card, model page, pricing page and documentation alongside the release.
Inside the model
Opus 4.1 is described as a hybrid reasoning model, meaning it can switch on extended thinking, which the company allows up to 64K tokens on some evaluations. Benchmark runs without extended thinking covered SWE-bench Verified and Terminal-Bench; TAU-bench, GPQA Diamond, MMMLU, MMMU and AIME were run with it. For agentic coding, the Claude 4 family uses a simple scaffold with two tools only: a bash tool and a string-replacement file editor. The planning tool from Claude 3.7 Sonnet was dropped. On TAU-bench, a prompt addendum told the model to write out its reasoning during multi-turn tool use, and the step limit rose from 30 to 100 completions.
What it can do
Anthropic reports 74.5% on SWE-bench Verified without extended thinking, which it calls its best coding result. The company also claims gains in research and data analysis, especially keeping track of fine detail and searching on its own. Outside feedback is cited: GitHub saw improvements across most capabilities, notably multi-file refactoring; Rakuten Group found it pinpointed exact fixes in large codebases without side effects; Windsurf measured a one standard deviation gain over Opus 4 on its junior developer benchmark. Claude 4 scores are reported over all 500 SWE-bench problems, while OpenAI figures use a 477-problem subset.
How to use it
The model is available to paid Claude users and inside Claude Code, and through the Anthropic API, Amazon Bedrock and Google Cloud's Vertex AI. Developers call it as claude-opus-4-1-20250805. Pricing matches Claude Opus 4. Anthropic publishes a system card, model page, pricing page and docs for it.
Reception and debate
Anthropic quotes three external teams. GitHub says most capabilities improved over Opus 4, with multi-file code refactoring standing out. Rakuten Group reports the model finds precise corrections in large codebases without needless edits or new bugs, and its team prefers that precision for daily debugging. Windsurf reports a one standard deviation improvement over Opus 4 on its junior developer benchmark, comparing the jump to the move from Sonnet 3.7 to Sonnet 4.
Why it matters
Anthropic recommends all users move from Opus 4 to Opus 4.1 and frames it as an incremental step, promising substantially larger model improvements soon. The release continues the Claude 4 approach of hybrid reasoning with a minimal two-tool agent scaffold, and it is offered through the same cloud channels as earlier Claude models.
Questions people ask about Claude Opus 4.1
What is Claude Opus 4.1?
It is Anthropic's updated version of Claude Opus 4, a hybrid reasoning language model aimed at agentic tasks, real coding and reasoning. The company reports 74.5% on SWE-bench Verified and says it improves research and data analysis work, especially detail tracking and agentic search.
How do I get it?
Paid Claude users and Claude Code have it, and developers can call it through the Anthropic API as claude-opus-4-1-20250805. It is also on Amazon Bedrock and Google Cloud's Vertex AI. Pricing is unchanged from Claude Opus 4.
What changed from Claude Opus 4?
Anthropic says the gains are in agentic tasks, real-world coding and reasoning. GitHub saw the biggest improvement in multi-file refactoring, Rakuten praised precise fixes in large codebases, and Windsurf measured a one standard deviation gain on its junior developer benchmark.
What does 'hybrid reasoning' mean here?
The model can use extended thinking, extra reasoning steps before answering. Some benchmark results were run without it, such as SWE-bench Verified, while others, including GPQA Diamond and AIME, used extended thinking of up to 64K tokens.
Good reading and sources
There is no English Wikipedia article on Claude Opus 4.1 yet, so the facts on this page come from the sources listed above: anthropic.com/news/claude-opus-4-1. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.