Claude Opus 5 vs GPT-5.6 Luna
| model Claude Opus 5 Anthropic | model GPT-5.6 Luna OpenAI | |
|---|---|---|
| Context | 1M | 1.1M |
| Input $/M | — | — |
| Output $/M | — | — |
| Knowledge cutoff | 2026-05 | 2026-02 |
| Available via | Anthropic API | OpenAI API |
| SWE-bench Verified (resolved) | 96 | — |
| SWE-bench Pro | 79.2% | 62.7% |
| SWE-bench Multilingual | 89.5% | — |
| SWE-bench Multimodal | 59.4% | — |
| DeepSWE | 68.8% | 67.2% |
| FrontierCode (mean@5) | 53.4 | — |
| Terminal-Bench (success rate) | — | 84.7 |
| GPQA Diamond (accuracy) | — | 92.3 |
| FrontierMath (accuracy) | — | 78.6 |
| BrowseComp (accuracy) | — | 83.3 |