Meet the New Claude Opus 5: Frontier-Class Agentic Coding and Computer Use at Unchanged Opus Pricing

Details

Anthropic today released Claude Opus 5, replacing Opus 4.8 as the Opus-tier flagship. Pricing holds at $5 per million input tokens and $25 per million output tokens — Anthropic positions this as near-Fable-5 intelligence at half the cost. It becomes the default model on Claude Max and the top option on Claude Pro. Context window is 1M tokens with 128k output on the synchronous Messages API and 300k via Message Batches. The headline benchmark numbers are decisive. On FrontierBench v0.1 — a 74-task agentic coding suite — Opus 5 scored 43.3% at max effort, up from Opus 4.8's 18.7% and above both GPT-5.6 Sol (37.5%) and Fable 5 (33.7%). SWE-bench Verified hit 96.0%. OSWorld 2.0 computer use climbed from 55.7% to 70.57%, and Zapier AutomationBench improved from 17.0% to 26.0%. Notably, Opus 5 safety classifiers flagged only 5% of API calls versus Fable 5's 42%, giving it a significant reliability advantage in production agentic pipelines. Three API-level changes demand immediate developer attention. First, thinking is now on by default — existing max_tokens values that assumed no thinking tokens will likely be too low. Second, setting thinking disabled alongside effort xhigh or max now returns a 400 error; integrations must either cap effort at high or remove the thinking field entirely. Third, the minimum cacheable prompt drops from 1,024 to 512 tokens, lowering prompt caching costs for shorter prompts. Anthropic also advises removing manual verification-step instructions from system prompts: Opus 5 self-verifies its work, and explicit instructions cause over-verification. The model ID is claude-opus-5; no smaller context variant exists.