OpenAI · GPT-6
GPT-6 Astra
OpenAI's top GPT-6 model for the hardest agentic, coding and research work, priced well above Sol and Luna.
- AA Intelligence
- 52.7#3 of 142
- Input / output per 1M
- $10.00/ $50.00
- Output speed
- 58tok/s
- Context window
- 1.05M tokens
Composite indexes
- 53
Leader: Claude Opus 5.5 at 58
Coding
- Terminal-Bench 4.0#3 of 8859.1%
Leader: Claude Mythos 5.1 at 60.9%
- Terminal-Bench 2.x#7 of 10388.4%
Leader: Claude Fable 5.1 at 91.4%
- SciCode#18 of 9456.5%
Leader: Claude Opus 5.5 at 66.9%
Agents and tool use
- GDPval-AA#17 of 1031542
Leader: Claude Opus 5.5 at 1846
- BrowseComp#1 of 1491.5%
Leader: GPT-6 Astra at 91.5%
- OSWorld-Verified#10 of 1672.6%
Leader: Kimi K3 at 84.8%
Reasoning and knowledge
- Humanity’s Last Exam#6 of 14754.7%
Leader: Claude Opus 5.5 at 61.4%
- GPQA Diamond#1 of 14296.1%
Leader: GPT-6 Astra at 96.1%
- CritPt#2 of 13931.7%
Leader: GPT-5.6 Sol at 32.3%
- ARC-AGI-2#1 of 2595%
Leader: GPT-6 Astra at 95%
- ARC-AGI-3#1 of 962.7%
Leader: GPT-6 Astra at 62.7%
Vision and long context
- AA Long Context Reasoning#25 of 13980.7%
Leader: Kimi K3 at 88.7%
- MMMU-Pro#2 of 7286.9%
Leader: Claude Opus 5.5 at 87.7%
Human preference
- LMArena Text#15 of 541480
Leader: Claude Fable 5 at 1506
- LMArena WebDev#2 of 541792
Leader: Claude Opus 5.5 at 1818
Other published results
| Result | Score | Reported by | Source |
|---|---|---|---|
| AA-Omniscience Index · max effort | 43.4 index (-100 to 100) | independent | Artificial Analysis: GPT-6 Astra (max) ↗ |
| AutomationBench-AA (partial score) · max effort | 68.5% | independent | Artificial Analysis: GPT-6 Astra (max) ↗ |
| Terminal-Bench Science (AA) · max effort | 63.3% | independent | Artificial Analysis: GPT-6 Astra (max) ↗ |
| AA Coding Agent Index v1.5 · Codex + GPT-6 Astra (max); DeepSWE 67.6 / SWE-Atlas-QnA 61.8 / TB4 55.6 | 61.6 index | independent | Artificial Analysis Coding Agents ↗ |
| Agents' Last Exam · max at any effort | 59.3% | vendor | OpenAI: GPT-6 Astra ↗ |
| DeepSWE v1.1 · max at any effort | 74.1% | vendor | OpenAI: GPT-6 Astra ↗ |
| Terminal-Bench 4.0 (vendor) · max at any effort | 57.9% | vendor | OpenAI: GPT-6 Astra ↗ |
| HLE (with tools) · max at any effort | 57.2% | vendor | OpenAI: GPT-6 Astra ↗ |
| FrontierMath Tier 4 (v2) · max at any effort | 97.6% | vendor | OpenAI: GPT-6 Astra ↗ |
| ScreenSpot-Pro (no tools) · max at any effort | 92.7% | vendor | OpenAI: GPT-6 Astra ↗ |
| Terminal-Bench Science 0.1 (vendor) · max at any effort | 64.6% | vendor | OpenAI: GPT-6 Astra ↗ |
| MRCR v2 8-needle 512K-1M · max at any effort | 96.3% | vendor | OpenAI: GPT-6 Astra ↗ |
| ARC-AGI-3 (Provider Adapter harness) · Max, OpenAI Provider Adapter harness | 98.55% | independent | ARC Prize: GPT-6 Astra ↗ |
| AA Intelligence Index (xhigh) · AA Intelligence Index v4.3.2, xhigh | 52.4 index | independent | Artificial Analysis: GPT-6 Astra (xhigh) ↗ |