GPT-6 Astra
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Math97.3
- 2Reasoning#193.3
- 3Multimodal87.3
- 4Agentic & tool use#786.5
- 5Knowledge#384.3
- 6Coding#476.4
Benchmark results (53)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| BrowseComp | 91.5 | #4 of 50 |
| MazeBench | 23 | #1 of 42 |
| HLE w/ tools | 57.2 | #10 of 27 |
| AutomationBench | 41.4 | #13 of 25 |
| OSWorld 2.0 | 72.6 | #1 of 24 |
| Terminal-Bench 4.0 | 57.90 | #4 of 18 |
| ApprenticeBench | 68 | #2 of 18 |
| Agents' Last Exam | 59.3 | #1 of 17 |
| Hermes Index | 56.3 | #3 of 16 |
| Hermes Bench | 73.7 | #4 of 16 |
| ExploitGym | 42.4 | #1 of 14 |
| CWE-bench v1 | 68.0 | #1 of 11 |
| Terminal-Bench-Science 0.1 | 64.6 | #1 of 5 |
Reasoning
| ARC-AGI-2 | 95.0 | #1 of 41 |
| ARC-AGI-1 | 97.50 | #4 of 36 |
| ARC-AGI-3 | 62.7 | #1 of 22 |
| STEMBench | 2.81 | #1 of 9 |
| ARC-AGI-3 Provider Adapter | 98.55 | #1 of 8 |
| GraphWalks BFS 256K–1M | 71.8 | #2 of 4 |
| GeneBench-Pro | 37.8 | #1 of 2 |
| MRCR v2 256K-512K | 100.0 | #1 of 2 |
| MRCR v2 512K-1M | 96.3 | #2 of 2 |
Coding
| DeepSWE | 74.1 | #8 of 50 |
| Bug Hunt Bench | 45.0 | #2 of 23 |
| FrontierSWE v2 | 65.5 | #1 of 20 |
| FrontierCode 1.1 Main | 53.3 | #4 of 16 |
| PostTrainBench v1.1 | 44.3 | #3 of 15 |
| FrontierCode 1.1 Extended | 64.5 | #1 of 8 |
Knowledge
| GPQA | 96.0 | #1 of 117 |
| HealthBench Professional | 64.7 | #4 of 16 |
| HealthBench Hard | 36.6 | #3 of 12 |
| HealthBench (raw) | 56.9 | #4 of 7 |
| HealthBench (length-adjusted) | 58.3 | #4 of 7 |
| HealthBench Professional (raw) | 68.2 | #4 of 7 |
Multimodal
| ParseBench | 70.67 | #11 of 37 |
| ExtractBench (LlamaIndex)(lower is better) | 91.91 | #19 of 24 |
| ScreenSpot Pro | 92.7 | #1 of 15 |
| ExtractBench Pro | 58 | #1 of 5 |
| BenchCAD Vision2Code (tools) | 0.959 | #3 of 4 |
Math
| FrontierMath v2 (Tiers 1-3) | 93.684 | #1 of 64 |
| FrontierMath v2 (Tier 4) | 97.600 | #2 of 51 |
Instruction following
| Gray Swan IPI (15 attempts)(lower is better) | 8.5 | #6 of 12 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| ✓ Official (OpenAI) checked Oct 17, 2026 · gpt-6-astra | $10.00 | $50.00 | $1.00 |
| BenchLM list price | $10.00 | $50.00 | $1.00 |
| OpenRouter openai/gpt-6-astra | $10.00 | $50.00 | $1.00 |
Standard tier, prompts under 272K tokens.
OpenAI's GPT-6 Astra model page lists $10.00 input / $1.00 cached input / $12.50 cache writes / $50.00 output per million tokens, a 1,050,000-token context window, 922,000 maximum input tokens, and 128,000 maximum output tokens. Prompts above 272K input tokens are charged at 2x input and cache rates and 1.5x output for the full request; Batch and Flex are 50% of Standard and Fast mode is 2x. The September 3, 2026 price sheet lists Astra at 2.5x GPT-5.6 Sol's current $4.00 / $20.00 rates.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text(max) | 1443 | #69 · 12.2K votes |
| WebDev (coding)(Max) | 1774 | #2 · 6.2K votes |
| Vision(Max) | 1280 | #36 · 4K votes |
| Search(max) | 1234 | #7 · 18.5K votes |
| Documents(max) | 1468 | #16 · 1.6K votes |
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Same family
- GPT-6 Sol78.6
- GPT-6 Luna66.6
Similar score
- GPT-6.1 Sol84.2 · $4.00
- Claude Fable 5.183.2 · $20.00
- Gemini 4 Argon82.9 · $4.00
- Claude Sonnet 5.582.2 · $4.00
- Claude Opus 5.586.7 · $8.00