Anthropic has released Claude Opus 5, its fourth model in a span of two months, and it surpasses Fable 5 on the majority of benchmarks.

      TL;DR: Claude Opus 5 exceeds Fable 5 on most performance metrics at half the token cost and is Anthropic's most aligned model so far.

      On Thursday, Anthropic unveiled Claude Opus 5, a model that matches or surpasses its flagship Fable 5 across most coding and knowledge-work assessments while costing just half the tokens. This is the fourth model released by Anthropic in less than two months, following Fable 5 in early June and Sonnet 5 at the end of June, with Opus 5 launching on July 24. Pricing is set at $5 per million input tokens and $25 per million output tokens, identical to its predecessor.

      In the Frontier-Bench coding evaluation, which assesses a model's ability to create functional software from engineering schematics, Opus 5 more than doubles the score of its predecessor and outperforms every competing model, including Fable 5. On ARC-AGI 3, which measures innovative problem-solving abilities, Opus 5 achieves approximately triple the score of the second-best model. In the Zapier AutomationBench, which evaluates models on their capability to execute real business tasks from start to finish, Opus 5 successfully completes more tasks than any competitor, even at its lowest effort setting.

      Anthropic also claims that Opus 5 is their most aligned model to date, showing the lowest incidence of deceptive behavior in its automated safety audit. While its predecessor set a new standard for honesty when it was released in late May, Opus 5 raises that standard further, albeit still trailing behind Mythos 5 in offensive cybersecurity tasks. Anthropic states that it purposely avoided training the model on cybersecurity tasks but has seen improvements in that area as a beneficial side effect of the model's general advancements.

      Opus 5 is now available as the new default on Claude Max and as the most powerful model on Claude Pro. Developers can access it via the API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. A Fast mode operates at two and a half times the default speed for double the base price, and two new API features are introduced: mid-conversation tool changes that enable developers to switch tools without disrupting the prompt cache, and automatic fallbacks that redirect flagged requests to a different model rather than blocking them.

      The speed of these releases highlights a wider trend in the AI model market. Anthropic has scaled up from one flagship model per quarter to four models in approximately eight weeks, each catering to different price-performance segments. Opus 5 serves as a bridge between the high-end Fable tier and the more affordable Sonnet tier, providing near-frontier intelligence for routine coding and knowledge tasks without the cybersecurity constraints of Fable or the access limitations of Mythos.