Claude Opus 5 brings near-Fable performance to the Opus tier

Anthropic released Claude Opus 5 on July 24, positioning it as a more practical frontier model: close to Claude Fable 5 in many areas, but priced at half the cost. The launch targets the gap between everyday use and the most demanding long-running agent work, where Anthropic still places Fable 5. 1 2

The model is available as claude-opus-5 through the Claude API and across Anthropic's platforms. It is the new default for Claude Max and the strongest model on Claude Pro. Anthropic also offers a Fast mode in research preview, which runs at about 2.5 times the default speed for twice the base price. 1

The list price is $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. Anthropic says that makes Opus 5 half the price of Fable 5 for comparable use. GitHub is already rolling the model into Copilot for Pro+, Max, Business, and Enterprise customers, although access is gradual and Business and Enterprise administrators must enable it. 1 3

Anthropic's published evaluations put Opus 5 ahead of other models on Frontier-Bench v0.1, near Fable 5's peak on CursorBench 3.2 at roughly half the task cost, and ahead on ARC-AGI 3, Zapier AutomationBench, and OSWorld 2.0. The company also reports gains over Opus 4.8 in life-science tasks, including 10.2 percentage points in organic chemistry and 7.7 points on protein-related work. These are launch-side results, not an independent audit, so the cost and ranking claims still need outside reproduction. 1

Opus 5 is not a blanket replacement for Fable 5. Anthropic says it remains weaker on long-running autonomous biology research and parts of offensive cybersecurity. Its classifiers can block binary vulnerability scanning, penetration testing, and exploit generation; GitHub likewise warns that some cyber and security-adjacent requests may be blocked. Claude products can fall back to Opus 4.8 when a request triggers a safety refusal. 1 2 3

The strategic change is therefore about model selection as much as raw capability. Anthropic is making an Opus-class system cheaper enough for routine coding and knowledge work while reserving Fable 5 for the longest, hardest agent runs. For developers, the immediate question is whether Opus 5's lower cost survives real workloads once independent evaluations and production usage data arrive.