Claude Opus 5 outperforms Fable 5 on agentic benchmark while cutting cost per task by 20%

Anthropic released Claude Opus 5 on July 24, 2026, positioning it as the new leader on Artificial Analysis’ agentic knowledge work benchmark, AA-Briefcase. Artificial Analysis reported that Opus 5 outperforms Fable 5 by nearly 150 Elo while reducing cost per task by 20%.

Public benchmark results from Epoch AI indicate mixed outcomes. Epoch AI’s ECI scores place Opus 5 at 159, slightly below Fable 5’s 161, while both models match at 161 on software engineering benchmarks (SWE-ECI).

Early user reports and anecdotes suggest Opus 5 delivers stronger real-world performance in coding and tool-use tasks than aggregate benchmarks currently reflect. One practitioner reported a clear head-to-head win over Fable 5, especially with best-of-n sampling, and highlighted effective browser automation capabilities. Another noted that Opus 5 successfully automated browser tasks such as canceling a ChatGPT Pro subscription.

Some evaluators observed irregular benchmark behavior for Opus 5. One noted that on FrontierCode, medium effort outperformed high effort, contrary to the typical pattern where more effort yields better results. This raises questions about evaluation stability or task-specific tradeoffs in inference-time compute.

Beyond capability claims, distribution details emerged. Nous Research’s portal added access to Opus 5 with a 20% discount applied to all models, including Opus 5, reflecting broader availability and pricing strategy.