GPT-5.4 mini scores 31.2 on the Respan Index (#63), ahead of Claude Sonnet 4.5 at 29.1 (#65). GPT-5.4 mini leads on 10 of the 13 benchmarks both report. GPT-5.4 mini is 3.6x cheaper per token at list price. GPT-5.4 mini has the larger context window (400K tokens).
* Estimated: no published scores in that category. How the Respan Index works
| Mystery Game Puzzles (Epoch AI) |
| 17% |
| 11% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 77.8% | 88.9% |
| SAGE (Vals AI) | 36.1% | 50.8% |
| SimpleQA Verified | 30.7% | 29.4% |
| WeirdML (Håvard Tveit Ihle) | 47.7% | 60.3% |
| LMArena Elo | 1457 | 1447 |
| LMArena WebDev (LMArena) | 1393 | 1397 |
| OSWorld-Verified | 61.4% | 72.1% |
| Terminal-Bench 2.0 | 51% | 60% |
Higher is better. Each score links to where it was published.