// in this post
Anthropic released Claude Opus 5 on Friday, pricing it at $5 per million input tokens and $25 per million output tokens, the same rates as its predecessor Opus 4.8 and half those of Claude Fable 5, the company's most capable widely released model.
The model is available immediately across the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry, under the identifier claude-opus-5. Anthropic has made it the default model on Claude Max subscriptions and the strongest model available on Claude Pro.
What Anthropic says it does
The company reported a series of benchmark results, all self-published and not independently verified at the time of writing. On Frontier-Bench v0.1, Anthropic said Opus 5 "more than doubles Opus 4.8's performance at a lower cost". On CursorBench 3.2, it said the model scored "within 0.5% of Fable 5's peak score, but at half the cost", and on OSWorld 2.0 it said Opus 5 surpassed "Fable 5's best result at just over a third of the cost".
Anthropic also said the model scored three times as high as the next-best model on ARC-AGI 3, and around 1.5 times the next-best pass rate on Zapier's AutomationBench. On scientific tasks, it reported gains of 10.2 percentage points over Opus 4.8 on organic chemistry and 7.7 percentage points on protein tasks. On legal work, the company said Opus 5 used 26% fewer tokens on average than Opus 4.8 at maximum reasoning.
According to Anthropic's platform documentation, Opus 5 has a context window of 1 million tokens, a maximum output of 128,000 tokens and a training data cutoff of May 2026, four months later than Fable 5's January 2026 cutoff.
Fast mode and new API features
Anthropic introduced a Fast mode that runs at 2.5 times the default speed for twice the base price, available through usage credits in Claude Code. Two features were released in beta: the ability to change tools mid-conversation, and automatic fallbacks on the API.
Safety evaluations and routing
Anthropic said evaluations conducted with private-sector and government partners found the model "does not advance the frontier in risky, dual-use capabilities". On cybersecurity, the company said Opus 5 "remains substantially behind Mythos 5 on the exploitation of those vulnerabilities". On biology, it said the model "still shows important limitations on long-running, autonomous research tasks, which is where we expect AI models to pose the most substantial biology-related risks".
The company said biology-related requests blocked on Fable 5 will now route to Opus 5 rather than Opus 4.8, while flagged cyber requests in Claude Code will fall back to Opus 4.8 by default.
Background
The launch follows Anthropic's move of Fable 5 to usage-based billing at $10 per million input tokens and $50 per million output tokens, double the rates of Opus 4.8, a change that took effect after several extensions. Anthropic's documentation continues to list Fable 5 as the model for workloads needing "the highest available capability", while recommending Opus 5 as the starting point for complex agentic coding and enterprise work.
Anthropic's documentation also states that Claude Opus 4.1 is deprecated and will be retired on 5 August 2026, directing users to migrate to Opus 5.