GPT-6 Astra
OpenAI · Released 3 Sep 2026 · gpt-6-astra
$10.00
Input price / 1M tokens
#59 of 61, cheapest first
$50.00
Output price / 1M tokens
#59 of 61, cheapest first
1.05M
Context window
#2 of 71
51
Output tokens / second
#52 of 61
57.9%
Terminal-Bench 4.0
#4 of 10
74.1%
DeepSWE v1.1
#4 of 17
72.6%
OSWorld 2.0
#3 of 10
91.5%
BrowseComp
#1 of 7
Summary
GPT-6 Astra is a proprietary model from OpenAI, released on 3 Sep 2026.
At $10.00 input and $50.00 output per million tokens, it is #59 of 61 on input price, cheapest first.
Its 1.05M-token context window ranks #2 of 71.
Measured output speed is 51 tokens per second, #52 of 61.
OpenAI's most capable model, for the hardest end-to-end work: complex reasoning, coding, computer use, research and document creation.
Benchmarks · 12 reported
| Benchmark | Score | Bar | Rank | Source |
|---|---|---|---|---|
| Terminal-Bench 4.0max score at any effort | 57.9% | #4 of 10 | OpenAI launch postSelf-reported | |
| DeepSWE v1.1max score at any effort | 74.1% | #4 of 17 | OpenAI launch postSelf-reported | |
| OSWorld 2.0v2026.08.08, offline set, partial score | 72.6% | #3 of 10 | OpenAI launch postSelf-reported | |
| Agents' Last Exammax score at any effort | 59.3% | – | OpenAI launch postSelf-reported | |
| BrowseCompmax score at any effort | 91.5% | #1 of 7 | OpenAI launch postSelf-reported | |
| AutomationBenchmax score at any effort | 41.4% | #1 of 8 | OpenAI launch postSelf-reported | |
| GPQA Diamondmax score at any effort | 96% | #1 of 27 | OpenAI launch postSelf-reported | |
| Humanity’s Last Exam (with tools)with tools | 57.2% | #9 of 16 | OpenAI launch postSelf-reported | |
| FrontierMath Tier 4 (v2)max score at any effort | 97.6% | – | OpenAI launch postSelf-reported | |
| ARC-AGI-2max score at any effort | 95% | #1 of 6 | OpenAI launch postSelf-reported | |
| ARC-AGI-3run with OpenAI's Responses API harness (two settings changed, per footnote 1) | 99.9% | – | OpenAI launch postSelf-reported | |
| ScreenSpot-Prono tools | 92.7% | – | OpenAI launch postSelf-reported |
Self-reported means the lab ran the test itself. A rank appears only where several labs report the same version of a benchmark.
Similar models
Models with the most benchmarks in common with GPT-6 Astra, and the closest scores.
Terminal-Bench 4.0
#4 of 10Long jobs in a real terminal
OSWorld 2.0
#3 of 10Long desktop and web workflows
- GPT-6 Astra72.6
- GPT-5.6 Sol62.6
- Claude Fable 5.177.9
- Claude Fable 572.9
- GPT-6.1 Sol71.4
Compare GPT-6 Astra side by side
Pre-filled with the two closest models. Swap any of them.
Sources
- price, batch/flex/fast/ultrafast tiers, long-context rule, data residency uplift: developers.openai.com/api/docs/pricing, checked 2 Oct 2026
- context, max output (max input 922K), knowledge cutoff, modalities, reasoning efforts, api id: developers.openai.com/api/docs/models/gpt-6-astra, checked 2 Oct 2026
- release date (API changelog, Sep 3 2026): developers.openai.com/api/docs/changelog, checked 2 Oct 2026
- cache write 1.25x, 30-minute TTL: developers.openai.com/api/docs/guides/prompt-caching, checked 2 Oct 2026
- benchmarks, launch details: openai.com/index/gpt-6-astra/, checked 2 Oct 2026
- output speed: artificialanalysis.ai/models/gpt-6-astra, checked 2 Oct 2026
Building on GPT-6 Astra?
Get the architecture right before the bill arrives.
We size caching, routing and fallbacks for your workload on a free call, and tell you where a cheaper model is good enough.