GPT-6 Astra scores 70.5 on the Respan Index (#2), ahead of Claude Opus 5 at 63.9 (#7). GPT-6 Astra leads on 24 of the 30 benchmarks both report. Claude Opus 5 is 2.0x cheaper per token at list price. GPT-6 Astra has the larger context window (1.1M tokens).
* Estimated: no published scores in that category. How the Respan Index works
| 73.2% |
| 97.6% |
| FrontierMath Tiers 1-3 v2 (Epoch AI) | 85.6% | 93.7% |
| FrontierSWE V2 (Proximal Labs) | 52% | 65.5% |
| Furniture Assembly (Epoch AI) | 60.8% | 80% |
| GPQA Diamond | 93.9% | 96% |
| LiveBench Agentic Coding (LiveBench) | 65.2% | 57.3% |
| LiveBench Coding (LiveBench) | 81.4% | 80.4% |
| LiveBench Data Analysis (LiveBench) | 74.6% | 83% |
| LiveBench Instruction Following (LiveBench) | 63.8% | 75.6% |
| LiveBench Language (LiveBench) | 88.7% | 89.4% |
| LiveBench Mathematics (LiveBench) | 95.7% | 96.8% |
| LiveBench Reasoning (LiveBench) | 91.2% | 92.7% |
| Mystery Game Puzzles (Epoch AI) | 59% | 84% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 98.9% | 100% |
| SAGE (Vals AI) | 49.4% | 46.4% |
| SimpleQA Verified | 59.9% | 75.6% |
| Terminal-Bench 4.0 (Vals AI) | 53.5% | 59.6% |
| Vending-Bench 2 (Andon Labs) | 11181.87 | 15514.7 |
| WeirdML (Håvard Tveit Ihle) | 91.8% | 93.6% |
| LMArena Agent (LMArena) | 0.0867 | 0.1227 |
| LMArena Elo | 1490 | 1477 |
| LMArena Vision (LMArena) | 1289 | 1284 |
| LMArena WebDev (LMArena) | 1695 | 1788 |
| BALROG (BALROG) | 63.4% | 68.3% |
| Terminal-Bench Science 0.1 | 29% | 64.6% |
| SimpleBench (SimpleBench) | 80.6% | 86.5% |
Higher is better. Each score links to where it was published.