Claude Opus 4.1

2025-08-05 · Publisher : Anthropic · Domain : Artificial Intelligence

Anthropic releases Claude Opus 4.1, an update to Opus 4 with better performance in coding, agents and reasoning, at the same price. The vendor recommends migrating from Opus 4 for all uses.

What changes

  • 74.5% on SWE-bench Verified, versus 72.5% for Opus 4.
  • Notable progress in multi-file refactoring (feedback from GitHub) and in search and detail tracking.
  • More precise fixes in large codebases, without unnecessary changes (feedback from Rakuten).
  • Windsurf measures a one standard deviation improvement on its junior developer benchmark.
  • Drop-in replacement for Opus 4, with no price change.

### Specifications - API identifier: claude-opus-4-1-20250805 (removed from the API on August 5, 2026, still available on Bedrock and Google Cloud according to the pricing page). - Availability: paid Claude users, Claude Code, Anthropic API, Amazon Bedrock, Google Cloud Vertex AI. - Context 200K and output 32K (secondary source). Knowledge cutoff not disclosed. - Some benchmarks use extended thinking with a budget of up to 64K tokens. ### Pricing | Input (USD / M tokens) | Output (USD / M tokens) | Cache write 5 min | Cache read | Batch | |---|---|---|---|---| | 15 | 75 | 18.75 | 1.50 | 7.50 / 37.50 | ### Benchmarks | Benchmark | Score | |---|---| | SWE-bench Verified (without extended thinking) | 74.5% | | Terminal-Bench, GPQA Diamond, MMMLU, MMMU, AIME | in an image table, figures not recorded | Scores published by the vendor. ### Key takeaways An incremental update at no extra cost, to be adopted as a replacement for Opus 4 for coding and agent uses.

Links

Tags : LLM, Anthropic, Claude, Opus, Codage, Agents