The new model costs $5 per million input tokens and $25 per million output tokens, matching the price of its predecessor, Opus 4.8
NurPhoto / Getty Images
Anthropic unveiled Claude Opus 5 on Friday, billing it as a model that comes close to matching Claude Fable 5's capabilities at half the price — $5 per million input tokens and $25 per million output tokens, the same rates charged for its predecessor, Opus 4.8.
Claude Opus 5 is rolling out across all of Anthropic's platforms, where it takes over as the default on Claude Max, the company's top consumer subscription, and claims the top spot among models offered on Claude Pro. Fable 5, launched in June, is priced at $10 per million input tokens and $50 per million output tokens.
On coding and knowledge work benchmarks including Frontier-Bench and GDPval-AA, Anthropic says Opus 5 sets new performance highs, though the company acknowledges it remains behind Mythos 5 on cybersecurity tasks. On Frontier-Bench v0.1, Opus 5 scored 43.3%, compared with Opus 4.8's 18.7% and Fable 5's 33.7%, the company said. On OSWorld 2.0, a computer-use benchmark, Anthropic says Opus 5 surpassed Fable 5's top result at roughly a third of the cost.
Anthropic is positioning the model as suited for daily professional use rather than the longest autonomous tasks, for which the company says Fable 5 remains the stronger choice. Customers can tune an effort dial built into the model, dialing back output quality in exchange for faster responses and reduced token consumption.
Early-access customers cited efficiency gains with unusual specificity. Niko Grupen, head of applied research at legal AI firm Harvey, reported that Opus 5 matched the output quality of Opus 4.8 running at maximum reasoning while cutting average token usage by 26%. Wade Foster, chief executive of Zapier, said the model completed a full churn-prevention workflow end to end on his company's AutomationBench — a task prior models failed — without spending more tokens than earlier Claude models.
According to Anthropic's internal behavioral audit, Opus 5 scored better on alignment measures than any prior model, including Opus 4.8, Sonnet 5, and Fable 5, showing the fewest instances of deceptive outputs and the greatest resistance to manipulation attempts. Anthropic said it deliberately withheld cyber-focused training from Opus 5 — a decision mirroring its approach with Opus 4.8 — yet the model's cybersecurity performance rose anyway, a byproduct of broader capability improvements. In Anthropic's OSS-Fuzz tests, Opus 5 found vulnerabilities 79.4% of the time — nearly level with Mythos 5's 80% — but when it came to turning those findings into working exploits, the model succeeded in just 4 of the challenges where Mythos 5 cleared 13.
Anthropic said Opus 5's safety classifiers are expected to intervene roughly 85% less often than those on Fable 5. When a classifier flags a request in Claude.ai, Claude Code, or Claude Cowork, the query falls back to Opus 4.8 by default, the company said.
The launch follows a turbulent stretch for Anthropic's top models. The U.S. government lifted export controls on Fable 5 and Mythos 5 earlier this month, ending an 18-day shutdown triggered by a government directive over cybersecurity concerns stemming from a technique Amazon $AMZN researchers documented for eliciting dangerous outputs from Fable 5.
Friday's release also brings several supporting features: a Fast mode that runs at about 2.5 times the standard speed for double the base price, automatic fallback routing built into the API, and support for swapping tools mid-conversation without busting the prompt cache. Developers can access the model as claude-opus-5 on the Claude API.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
