Overview
- On August 1, Anthropic revoked OpenAI’s general access to its Claude API, citing a terms-of-service breach for using the model to develop GPT-5.
- The company said it will preserve limited API use for benchmarking and safety evaluations despite removing broader access.
- OpenAI’s chief communications officer called peer benchmarking industry standard and lamented that Anthropic retains full access to OpenAI’s API even as OpenAI has been cut off.
- Claude Opus 4.1 went live August 5 on Claude.ai, Anthropic’s API, Amazon Bedrock and Google Cloud’s Vertex AI with unchanged pricing.
- The new model scored 74.5% on the SWE-bench Verified benchmark—outperforming OpenAI’s o3 at 69.1%—and drew praise for clean multi-file code refactoring from GitHub and Rakuten.