Anthropic Ships Claude Opus 5: Nearly Fable at Half the Price

Anthropic released Claude Opus 5 on 24 July. It is the fourth Claude 5 release in two months, following Mythos 5, Fable 5 and Sonnet 5. Opus 5 sits between flagship Fable 5 and the lighter Sonnet 5: performance close to the flagship, at half the price.

Has the price changed, and who gets access

The price holds steady: $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. The model becomes the default on the Claude Max plan and the strongest option on Claude Pro. For the most demanding autonomous work, Anthropic still recommends Fable 5.

How does it score against the flagship

Opus 5 outperforms its rivals on the benchmarks. On Frontier-Bench v0.1 (software engineering tasks) it sets a new industry high, doubling Opus 4.8 at a lower cost per task. On CursorBench 3.2 it lands within 0.5% of the peak Fable 5 result at half the price. On ARC-AGI 3 it leads the nearest competitor threefold. On OSWorld 2.0 (a computer use benchmark) Opus 5 beats Fable 5 at a third of the cost.

What changes for data retention and classifiers

One notable difference from Fable 5: the 30-day data retention rule does not apply to Opus 5. Restrictive classifiers fire 85% less often. For the cases where a classifier does block a request, a beta feature called Automatic Fallbacks has been launched — it reroutes the request to a less powerful model instead of returning an error.

What does this mean for the budget

The model is already available through the API and the consumer plans. For Russian companies that use Claude through API providers, model pricing and limits feed directly into the budget. It is worth recalculating what an AI rollout costs with the new model in mind: for most business tasks, the price-to-quality ratio of Opus 5 now beats the flagship.