Anthropic announced Claude Opus 5 on July 24, 2026, positioning it as a daily-driver flagship that approaches the frontier intelligence of Claude Fable 5 at half the price. Per the launch page, Opus 5 ships with a 1M-token context, costs $5 per million input tokens and $25 per million output tokens (matching its predec The model introduces an effort dial that lets customers trade intelligence against token cost, and Anthropic reports it as state-of-the-art on coding and knowledge-work evaluations including Frontier-Bench v0.1, CursorBench 3.2, ARC-AGI 3, OSWorld 2.0, GDPval-AA v2, HLE, and Zapier AutomationBench, while still trailing

Claude Platform release notes dated September 1, 2026 confirm that Claude Opus 5 is part of the beta for per-message effort changes on the Claude API, letting callers adjust effort for later turns inside a conversation via a system message carrying output_config.effort. The same notes describe thinking-block interopera For accounts created on or after August 31, 2026, replaying a thinking block after the system prompt, tools, or an earlier message has changed returns a 400 error, with a thinking-binding-controls-2026-08-01 beta header available to control whether mismatched blocks are rejected or dropped. On Fable 5.1 and Mythos 5.1,