Anthropic has launched a new iteration of its AI model as OpenAI prepares to release GPT-5. The new Anthropic model will be available to paid Claude users and in Claude Code. OpenAI's GPT-5 is expected to launch in early August, unifying the company's GPT and O series of AI models. GPT-5 is anticipated to show improvements in software engineering, performing better than Claude Sonnet 4.
Anthropic has restricted OpenAI's access to Claude's API, alleging that OpenAI's engineers were utilising it to fine-tune GPT-5. Anthropic's terms of service forbid using the service to build a competing product or reverse engineer the service. However, Anthropic will grant OpenAI access for benchmarking and safety evaluations.
Claude 4 currently leads in coding benchmarks, outperforming OpenAI and Google models. On Anthropic's SWE-Bench, Claude Opus 4 scored 72.5%, the highest of any model tested. It excels at writing correct code, debugging, and iterating on its output.
