GPT-6 Astra scores 70.5 on the Respan Index (#2), ahead of DeepSeek V4 Pro at 46.9 (#29). GPT-6 Astra leads on 23 of the 23 benchmarks both report. DeepSeek V4 Pro is 10x 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
| 45.3% |
| 93.7% |
| GPQA Diamond | 90.9% | 96% |
| LiveBench Agentic Coding (LiveBench) | 54.9% | 57.3% |
| LiveBench Coding (LiveBench) | 77.2% | 80.4% |
| LiveBench Data Analysis (LiveBench) | 79.2% | 83% |
| LiveBench Instruction Following (LiveBench) | 67.7% | 75.6% |
| LiveBench Language (LiveBench) | 82.1% | 89.4% |
| LiveBench Mathematics (LiveBench) | 95.1% | 96.8% |
| LiveBench Reasoning (LiveBench) | 85.8% | 92.7% |
| Mystery Game Puzzles (Epoch AI) | 43% | 84% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 96.7% | 100% |
| SimpleQA Verified | 47% | 75.6% |
| Terminal-Bench 4.0 (Vals AI) | 14.1% | 59.6% |
| Vending-Bench 2 (Andon Labs) | 3284.52 | 15514.7 |
| WeirdML (Håvard Tveit Ihle) | 66.2% | 93.6% |
| LMArena Agent (LMArena) | 0.0119 | 0.1227 |
| LMArena Elo | 1451 | 1477 |
| LMArena WebDev (LMArena) | 1583 | 1788 |
| SimpleBench (SimpleBench) | 50.9% | 86.5% |
Higher is better. Each score links to where it was published.