Anthropic launches Claude Opus 5: near-Fable performance at half the price, effort toggle for cost control

Anthropic released Claude Opus 5 on July 24, 2026, as a near-frontier model priced at $5 per million input tokens and $25 per million output tokens—identical to Opus 4.8 and half of Fable 5's $10/$50. The new model is Anthropic's fourth major release in under two months (following Mythos 5, Fable 5, and Sonnet 5 in June) and is now the default on Claude Max. Opus 5 carries a 1-million-token context window, 128k max output, and a new 'effort' toggle (low, medium, high, xhigh, max) enabling users to trade reasoning depth for token efficiency on a per-request basis.

On Frontier-Bench v0.1, Opus 5 scores 43.3% at maximum effort versus Fable 5's 33.7%, achieving near-frontier performance on coding and knowledge-work benchmarks while consuming fewer tokens on average. Enterprise early-access customers reported concrete wins: Harvey (legal AI) saw 26% fewer tokens at max reasoning, Zapier completed end-to-end churn-prevention sequences previously unsolved, and Cursor confirmed near-Fable coding speeds with significantly faster execution. The model emphasizes token efficiency for enterprise customers concerned about runaway AI bills, and includes stronger alignment and less restrictive cybersecurity classifiers than Fable (85% lower refusal trigger sensitivity, with automatic fallbacks instead of hard errors).

Opus 5's timing aligns with week-of-release pressure from competing models: Moonshot's Kimi K3 topped coding leaderboards ahead of Fable 5, while Fable 5's free-tier access closed July 19 amid capacity strain. By delivering near-flagship performance at unchanged Opus pricing with granular effort controls, Anthropic is asserting a clear cost/capability floor below Fable 5—positioning Opus 5 as the workhorse tier for teams managing inference budgets while retaining access to frontier-class reasoning when needed.