GPT-6 Astra
OpenAI · Flagship · Sep 3, 2026New
OpenAI's flagship. It leads the independent reasoning tests in this catalogue: ARC-AGI-2, ARC-AGI-3 and Humanity's Last Exam.
- Artificial Analysis Intelligence Index
- 52.7#4 of 29 · max effort
- API price per 1M tokens
- $10 / $50input / output · $20 blended
- Context window
- 1.1Mup to 128K output tokens
- Output speed
- 48 t/s6m 54s to first answer at max effort
Benchmark results
Each row shows this model (blue) against every other model in the catalogue (grey), worst on the left, best on the right.
Overall capability
- Artificial Analysis Intelligence Index52.7#4 of 29
Max effort
- Epoch Capabilities Index166.4#2 of 16
- LiveBench82.2%#3 of 23
Max effort
Coding
- Terminal-Bench 4.058.2%#3 of 13
Codex harness, max effort
- SWE-Bench Pro (V2 Hard)90.2%#2 of 5
Codex harness, high effort
- Arena WebDev1,788#2 of 27
Max effort
Agents & professional work
- GDPval-AA1,542#14 of 18
Max effort
- Vending-Bench 2$15,515#1 of 9
Reasoning
- ARC-AGI-295%#1 of 11
Max effort
- ARC-AGI-362.7%#1 of 5
Max effort
- FrontierMath (Tiers 1–3)93.7%#1 of 14
Max effort
- MathArena88%#2 of 10
Max effort
- SimpleBench83.6%#3 of 12
Knowledge & science
- Humanity's Last Exam (Diamond)60.6%#1 of 9
- GPQA Diamond95.8%#1 of 14
Max effort
Multimodal
- Arena Vision1,284#6 of 15
Max effort
Human preference
- Arena Text1,477#12 of 24
Max effort
Reasoning effort trade-offs
The same model at different thinking budgets. More effort buys a higher intelligence score at a higher cost per task and a longer wait. Measured by Artificial Analysis.
| Effort | Score | Cost / task | First answer | Speed |
|---|---|---|---|---|
| Low | 45.8 | $0.82 | 3s | 46 t/s |
| Medium | 49.6 | $1.54 | 5.6s | 46 t/s |
| High | 50.9 | $1.73 | 1m 35s | 47 t/s |
| Xhigh | 52.4 | $2.31 | 3m 47s | 46 t/s |
| Max (headline score) | 52.7 | $3.26 | 6m 54s | 48 t/s |
Specs & pricing
- Developer
- OpenAI
- Released
- Sep 3, 2026
- Availability
- Generally available
- API model ID
gpt-6-astra- Weights
- Proprietary (API only)
- Reasoning
- Always on
- Knowledge cutoff
- Apr 2026
- Accepts
- Text, Image
- Produces
- Text
- Context window
- 1,050,000 tokens
- Max output
- 128,000 tokens
- Input price
- $10 / 1M tokens
- Output price
- $50 / 1M tokens
- Cached input
- $1 / 1M tokens
Prompts above 272K input tokens are billed at twice the input price and 1.5 times the output price for the whole request.
Closest alternatives
Models from other labs nearest to GPT-6 Astra on Artificial Analysis Intelligence Index.
Gemini 4 Argon
Google
Intelligence 52.6 · $2 / $10
Compare side by sideClaude Fable 5.1
Anthropic
Intelligence 53.4 · $10 / $50
Compare side by sideClaude Sonnet 5.5
Anthropic
Intelligence 56 · $2 / $10
Compare side by side