OpenAI
availableShows if the model has enough results for an index.GPT-6 Astra
GPT-6 Astra is a reasoning model from OpenAI in the GPT-6 family. 48 benchmarks count toward its score, in 7 categories.
IndexOverall score. 50 is the middle.83.5 ±2.8
CoverageShare of the index weight with results.95%
SpeedOutput tokens per second.28/s
Input / 1MUS dollars per 1M input tokens.$10 batch $5
Output / 1MUS dollars per 1M output tokens.$50 batch $25 US dollars per 1M output tokens in a batch.
ContextMaximum tokens in one request.1.05M
EloLMArena rating and rank.1444 (#59)
50 is the middle of the board. The range shows the doubt in the index. Batch work costs less.
2,693 votes. Elo shows what people prefer. It does not change the score.
CapabilitiesScore per category. 50 is the middle.
50 is the middleResults
48 counted| BenchmarkThe test name. | CategoryThe capability that the test measures. | ResultThe score from the publisher. | IndexThis result on the index scale. | RunThe settings of the run. | DateDate of the result. | Published byThe source of the result. |
|---|---|---|---|---|---|---|
| OpenAI MRCR v2 8-needle 256K-512K | Reasoning | 100.0% | — | — | — | Meta Superintelligence Labs |
| IOI | Coding | 100.0% | 89.2 | — | 21 Sept 2026 | Vals AI |
| OTIS Mock AIME 2024-2025 | Math | 100.0% | 67.4 | max effort | — | Epoch AI |
| ProofBench v1.1 | Math | 99.0% | 89.5 | — | 21 Sept 2026 | Vals AI |
| FrontierMath-Tier-4-v2-Private | Math | 97.6% | 94.8 | high effort | — | Epoch AI |
| ARC-AGI-1 (semi-private) | Reasoning | 97.5% | 73.2 | max effort | — | ARC Prize Foundation |
| OpenAI MRCR v2 8-needle 512K-1M | Reasoning | 96.3% | — | — | — | Meta Superintelligence Labs |
| Artificial Analysis GPQA Diamond | Knowledge | 96.1% | 67.0 | — | — | Artificial Analysis |
| Graduate-Level Google-Proof Q&A | Knowledge | 96.0% | 66.9 | — | — | David Rein et al. |
| GPQA Diamond | Knowledge | 96.0% | 66.9 | — | — | David Rein et al. |
| BenchCAD Vision2Code voxel IoU with tools | Multimodal | 95.9% | — | — | — | Zhang et al. and Anthropic |
| GPQA diamond | Knowledge | 95.8% | 66.7 | max effort | — | Epoch AI |
| ARC-AGI-2 (semi-private) | Reasoning | 95.0% | 87.6 | max effort | — | ARC Prize Foundation |
| FrontierMath-Tiers-1-3-v2-Private | Math | 93.7% | 84.2 | max effort | — | Epoch AI |
| ScreenSpot Pro | Multimodal | 92.7% | 75.1 | — | — | Kaixin Li et al. |
| BrowseComp | Agentic | 91.5% | 76.6 | — | — | OpenAI |
| Vibe Code Bench v1.1 | Coding | 89.6% | 79.4 | OpenHands | 21 Sept 2026 | Vals AI |
| Terminal-Bench 2.1 | Agentic | 87.3% | 75.6 | — | 21 Sept 2026 | Vals AI |
| Artificial Analysis MMMU-Pro | Multimodal | 86.9% | 73.3 | — | — | Artificial Analysis |
| Mystery Game Puzzles | Reasoning | 84.0% | 95.0 | max effort | — | Epoch AI |
| Artificial Analysis Long Context Reasoning | Reasoning | 80.7% | 64.1 | — | — | Artificial Analysis |
| Furniture Assembly | Reasoning | 80.0% | 95.0 | max effort | — | Epoch AI |
| EuroEval Italian | Multilingual | 77.8% | 95.0 | — | — | EuroEval |
| EuroEval Swedish | Multilingual | 77.6% | 95.0 | — | — | EuroEval |
| Artificial Analysis Coding Index | Coding | 76.9% | 73.2 | — | — | Artificial Analysis |
| EuroEval French | Multilingual | 76.3% | 95.0 | — | — | EuroEval |
| EBR-bench | Reasoning | 76.2% | 95.0 | max effort | — | Epoch AI |
| SimpleQA Verified | Knowledge | 75.6% | 91.6 | max effort | — | Epoch AI |
| EuroEval Portuguese | Multilingual | 74.7% | 95.0 | — | — | EuroEval |
| EuroEval Dutch | Multilingual | 74.7% | 95.0 | — | — | EuroEval |
| DeepSWE | Agentic | 74.1% | 77.6 | — | — | Datacurve AI |
| OSWorld 2.0 | Agentic | 72.6% | 90.5 | — | — | Mengqi Yuan et al. |
| Chess Puzzles | Reasoning | 72.0% | 95.0 | max effort | — | Epoch AI |
| EuroEval Spanish | Multilingual | 70.0% | 89.7 | — | — | EuroEval |
| Artificial Analysis AutomationBench | Agentic | 68.5% | 83.1 | — | — | Artificial Analysis |
| HealthBench Professional raw score | Knowledge | 68.2% | — | — | — | Anthropic |
| EuroEval German | Multilingual | 68.0% | 87.2 | — | — | EuroEval |
| ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable job | Agentic | 68.0% | 95.0 | — | — | NeoCognition |
| Code Migration | Coding | 67.7% | 88.3 | — | 21 Sept 2026 | Vals AI |
| EuroEval Polish | Multilingual | 67.4% | 86.4 | — | — | EuroEval |
| FrontierSWE v2 | Coding | 65.5% | 90.6 | — | — | Proximal |
| HealthBench Professional | Knowledge | 64.7% | — | — | — | Rebecca Soskin Hicks et al. |
| Terminal-Bench-Science 0.1 | Agentic | 64.6% | — | — | — | Terminal-Bench-Science Team |
| FrontierCode 1.1 Extended | Coding | 64.5% | — | — | — | Cognition |
| ARC-AGI-3 (semi-private) | Reasoning | 62.7% | — | max effort | — | ARC Prize Foundation |
| Artificial Analysis Omniscience Accuracy | Knowledge | 62.6% | 91.3 | — | — | Artificial Analysis |
| Agents' Last Exam | Agentic | 59.3% | 95.0 | — | — | DeepSeek-AI |
| HealthBench length-adjusted score | Knowledge | 58.3% | — | — | — | Anthropic |
| Terminal-Bench 4.0.0 | Agentic | 58.2% | 95.0 | max effort · Codex | 21 Sept 2026 | Terminal-Bench |
| Humanity's Last Exam with tools | Agentic | 57.2% | 70.0 | — | — | DeepSeek-AI |
| Terminal-Bench 4.0 | Agentic | 57.1% | 95.0 | — | 21 Sept 2026 | Vals AI |
| HealthBench raw score | Knowledge | 56.9% | — | — | — | Anthropic |
| Artificial Analysis SciCode | Coding | 56.5% | 71.3 | — | — | Artificial Analysis |
| Artificial Analysis Humanity's Last Exam | Knowledge | 54.7% | 83.9 | — | — | Artificial Analysis |
| FrontierCode 1.1 Main | Coding | 53.3% | 81.0 | — | — | Cognition |
| Artificial Analysis Intelligence Index | Knowledge | 52.7% | 88.3 | — | — | Artificial Analysis |
| GDPval-AA normalized | Agentic | 52.1% | 75.6 | — | — | Artificial Analysis |
| Artificial Analysis Agentic Index | Agentic | 51.5% | 79.1 | — | — | Artificial Analysis |
| Artificial Analysis AnalystAgent | Agentic | 51.2% | 77.4 | — | — | Artificial Analysis |
| MirrorCode | Coding | 46.7% | — | high effort | — | Epoch AI |
| ExploitGym | Agentic | 42.4% | 91.5 | — | — | Zhun Wang et al. |
| AutomationBench | Agentic | 41.4% | 84.8 | — | — | Moonshot AI |
| Artificial Analysis Tau3-Banking | Agentic | 41.4% | 68.9 | — | — | Artificial Analysis |
| GeneBench-Pro | Reasoning | 37.8% | — | — | — | OpenAI |
| HealthBench Hard | Knowledge | 36.6% | 78.3 | — | — | Meta AI |
| Medical Long Context Reasoning (MLCR-AA) | Reasoning | 35.0% | 68.3 | — | — | Wisedocs and Artificial Analysis |
| Critical Physics Tasks | Reasoning | 31.7% | 95.0 | — | — | Artificial Analysis |
| Artificial Analysis GDP.pdf | Agentic | 31.0% | 82.8 | — | — | Artificial Analysis |
| Vibe Code Bench 1-100 | Coding | 27.6% | 82.2 | OpenHands | 16 Sept 2026 | Vals AI |
| Agent Arena task outcome | Agentic | 17.7 | 87.3 | max effort | 15 Sept 2026 | LMArena |
| Agent Arena command recovery | Agentic | 7.3 | 75.6 | max effort | 15 Sept 2026 | LMArena |
| ProgramBench | Coding | 5.5% | — | — | 21 Sept 2026 | Vals AI |
| FrontierMath-Erdos | Math | 2.9% | — | max effort | — | Epoch AI |
| Agent Arena steerability | Agentic | -0.5 | 66.8 | max effort | 15 Sept 2026 | LMArena |
48 benchmarks count, from 60 of 74 results. A grey row does not count. Too few models took that benchmark.