Claude Opus 5 5

Anthropic's release post positions Opus 5.5 as a major step up in agentic coding and computer use, priced 40% below Opus 5 and running more than 30% faster, with its strongest behavioral-audit scores to date and a one-time reset of Claude Code usage limits. Artificial Analysis independently placed it first on its Intelligence Index at 58, level with GPT-6 Astra on Terminal-Bench 4.0 and ahead on agentic knowledge-work evals. The practical catch sits in the integration notes. Vercel's AI Gateway announcement says the model retires forced tool-calling and fixed thinking budgets, so existing requests that disable thinking, pin a budget, or force a specific tool will fail with HTTP 400 errors. Teams should audit those request shapes before flipping the model string. Anthropic's developer team also published usage advice: hand over whole tasks with clear criteria for done, drop prompts telling the model to think carefully because it already reasons first, and check in after long autonomous runs rather than micromanaging each step.