GPT-6 Astra
Previous flagship. September’s flagship, still sold; it powers ChatGPT Pro reasoning.
Which reasoning effort setting?
controls how long GPT-6 Astra thinks before answering. Token prices never change; higher settings spend more and raise task cost.
Start at Xhigh. It scores 55.3 against 56.9 at the top setting, for 41% less per benchmark task. Rule: cheapest setting within 2 points of the model’s best score.
At Xhigh, GPT-6 Astra thinks for about 14.7K tokens per task, then writes about 12K more as its answer: $3.66 per task, 2.6× Low, for +12.0 points.
Output rate: $50.00/1M at every effort level. Thinking tokens are billed as output.
| Effort | Index | Per task | Thinking | All output | × Low | Time |
|---|---|---|---|---|---|---|
| Low | 43.3 | $1.42 | 3.3K | 9.1K | 1.00× | — |
| Medium | 51.6 | $2.73 | 8.6K | 18.1K | 1.92× | — |
| High | 54.8 | $3.24 | 12.1K | 23.1K | 2.27× | — |
| XhighStart here | 55.3 | $3.66 | 14.7K | 26.7K | 2.57× | — |
| Max | 56.9 | $6.25 | 35.8K | 52K | 4.39× | — |
Per task of the Artificial Analysis Intelligence Index: the average over its nine evaluations, counting input, cache, and output tokens at list prices. These are long, agent-style benchmark tasks, much longer than a chat message.
Where GPT-6 Astra sits on the frontier
Intelligence Index against one 100K-input, 10K-output request.