claude-opus-5-thinking

Sonar published an independent evaluation of Claude Opus 5 Thinking (adaptive thinking mode) against a Claude Opus 4.8 Thinking baseline using its LLM evaluation framework on a Java benchmark of 4,441 tasks (158 HumanEval, 386 MBPP, and 3,897 ComplexCodeEval), analyzed with SonarQube. The review reports an 88.6% pass r Beyond correctness, Sonar found that comment-line density rose from 6.9% to 10.5%, cognitive complexity per kLOC fell 10%, and Opus 5 generated 2.3 times more code than Opus 4.8, while total findings rose 2.7 times and code smell and overall issue density increased. The page notes that Anthropic describes Opus 5 as del

Apito's release guide consolidates Anthropic's July 24, 2026 announcement and Claude Platform documentation, confirming the official API model ID as claude-opus-5 with a 1M token context window, 128k max output, and adaptive thinking on by default. It lists effort levels as low, medium, high, xhigh, and max, and states The guide frames Opus 5 as a high-end default for many real developer workflows by matching Opus 4.8's standard API price while approaching Fable 5-class performance on coding, agent, automation, and knowledge-work tasks, and it cautions that max tokens must now cover both thinking and the visible final response, which

Anthropic's official Claude Platform documentation confirms Claude Opus 5 as a new model with API model ID claude-opus-5, a 1M token context window (1M is both default and maximum, with no smaller variant), 128k max output tokens, and thinking enabled by default. The page notes that Priority Tier is not supported on Cl The release notes introduce several new features: mid-conversation tool changes (beta) that let developers add or remove tools between turns while preserving the prompt cache, a new "default" mode for the fallbacks parameter that uses Anthropic's recommended fallback models by refusal category, and a lower prompt cache