Key facts
- Anthropic released Claude Opus 5 on July 24.
- Opus 5 costs $5 per million input tokens and $25 per million output tokens.
- Opus 5 outperforms Fable 5 and GPT-5.6 Sol on several key benchmarks.
- The new model has a less restrictive data retention policy than previous models.
- Anthropic is introducing a beta feature called Automatic Fallbacks.
Anthropic has released Claude Opus 5, a new AI model priced at $5 per million input tokens, which is half the cost of its predecessor, Fable 5. Released on July 24, Opus 5 also outperforms Fable 5 on most major benchmarks, including Frontier-Bench v0.1 and ARC-AGI-3, and is now the default model for Claude Max and the strongest option on Claude Pro.
Opus 5 achieved a 43.3% score on Frontier-Bench v0.1, a benchmark for AI coding agents, surpassing Fable 5's 33.7% and OpenAI's GPT-5.6 Sol's 34.4%. On the ARC-AGI-3 problem-solving benchmark, Opus 5 scored 30.2%, significantly higher than GPT-5.6 Sol's 7.8%. On the GDPval-AA v2 knowledge work benchmark, Opus 5 reached an Elo rating of 1,861, compared to Fable 5's 1,747 and GPT-5.6 Sol's 1,736. Zapier reported Opus 5 achieved 100% on its AutomationBench evaluation for business workflows.
Anthropic positions Opus 5 as a more cost-effective and less restrictive alternative to Fable 5. Unlike Fable and Mythos models, Opus 5 does not have a 30-day data retention policy, addressing privacy concerns. While Opus 5 includes safeguards, particularly for cybersecurity tasks, they are expected to engage approximately 85% less often than with Fable 5. Anthropic is also introducing a beta feature called Automatic Fallbacks, which reroutes prompts triggering safety classifiers to a less powerful model to provide a functional response.
The release comes a week after Beijing-based Moonshot AI unveiled Kimi K3, a 2.8-trillion-parameter open-weight model. Independent benchmarks place Kimi K3 third overall, behind