Anthropic has released Claude Opus 5, a new frontier model that has triggered significant discussion regarding benchmark scores and practical performance. Independent evaluations from Epoch AI report an Epoch Capabilities Index (ECI) of 159 and a SWE-ECI of 161 for the new model.
- Claude Opus 5 achieves an ECI of 159, slightly below Fable 5's score of 161.
- The model matches Fable 5 on software engineering benchmarks with a SWE-ECI of 161.
- Community feedback suggests the ECI score understates Opus 5's practical improvements over Opus 4.8, which scored only one point lower.
- Some evaluators noted non-monotonic performance on FrontierCode, where medium-effort inputs outperformed high-effort ones.
- Nous Portal has made the model available with a 20% discount on all models.
The launch highlights ongoing debates about the sensitivity of current evaluation metrics and the gap between aggregate scores and real-world coding agent capabilities.