Anthropic launched Claude Opus 5 on Friday, positioning the model as a more efficient option for advanced coding, AI agents and professional knowledge work.

Anthropic said Opus 5 comes close to Claude Fable 5’s performance at half the cost, including on CursorBench 3.2, where it finished within 0.5% of Fable 5’s peak score at half the cost per task.

Opus 5 is now available across Anthropic’s platforms and through the Claude API. It is the new default model for Claude Max subscribers and the company’s strongest model available through Claude Pro.

The model is priced at $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.8 and half the price of Claude Fable 5. Developers can access it through the API using the claude opus 5 model name.

Anthropic said Opus 5 set new state of the art results on coding and knowledge work evaluations including Frontier Bench and GDPval AA. On Frontier Bench v0.1, the model more than doubled the performance of Opus 4.8 while carrying a lower cost per completed task.

At its maximum effort setting, Opus 5 finished within 0.5% of Fable 5’s highest CursorBench 3.2 score while costing half as much per task, according to the company. Anthropic also said Opus 5 exceeded Fable 5’s best OSWorld 2.0 computer use result at slightly more than one third of the cost.

The model also recorded a score three times higher than the next strongest model on ARC AGI 3 and achieved a pass rate around 1.5 times higher than its nearest competitor on Zapier’s AutomationBench at the same cost per task, Anthropic said. The results were published by Anthropic and should be viewed within the methodologies and agent setups used for each evaluation.

Anthropic said Opus 5 is more likely to verify its work, identify underlying problems and continue iterating until a task succeeds. In one evaluation, the model created its own computer vision pipeline to extract geometry from an image before rebuilding the object as a three dimensional FreeCAD model.

The company also reported improvements across its life sciences evaluations. Opus 5 scored 10.2 percentage points above Opus 4.8 on an internal organic chemistry benchmark and 7.7 points higher on a protein function task.

Opus 5 remains behind Anthropic’s restricted Mythos 5 model in biology research and offensive cybersecurity. Anthropic said the two models showed similar success in identifying software vulnerabilities, but Opus 5 performed considerably worse when asked to develop exploits.

The model includes safeguards that block penetration testing, binary vulnerability scanning and exploit generation. Anthropic expects those classifiers to intervene around 85% less frequently than the safeguards applied to Fable 5, with flagged requests routed to Opus 4.8 by default in its consumer products.

Anthropic is also offering a Fast mode that runs around 2.5 times faster than the standard version at twice the base price. The launch includes beta support for changing tools during an active API conversation and automatically routing flagged requests to another Claude model.