Claude Opus 4.1

Released August 5, 2025 as a drop-in upgrade over Claude Opus 4, Opus 4.1 targeted agentic tasks, real-world coding, and reasoning with measurable gains: 74.5% on SWE-bench Verified and 43.3% on Terminal-Bench (ahead of Opus 4, o3, and Gemini 2.5 Pro on the latter). It is a hybrid reasoning model with extended thinking that can use tools mid-thought, excelling at multi-file refactoring, precise debugging, and hours-long autonomous operation. Anthropic positioned it as the model to upgrade to for all Opus 4 uses, and it shipped across the Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI.\n\nLimitations: premium pricing ($15/1M input, $75/1M output) makes it the most expensive mainstream option per token. It was quickly superseded within months by the 4.5 generation. Like all Anthropic flagships it is closed-weights, so no self-hosting or fine-tuning of the base model.

Anthropic-reported: 74.5% SWE-bench Verified, 43.3% Terminal-Bench