Model Release Radar
Price/capability frontier position, benchmark scores, and community signal
Model Release Radar
GPT-6 Astra
openai·released Thursday, Sep 3, 2026
Not enough paired price and capability data exists yet to place this model on the price/capability frontier.
This model has not been measured on DeepSWE (or any other benchmark with a real measured per-task cost) yet, so no cost/capability chart is drawn. Its Artificial Analysis benchmark scores are listed below.
Measured cost per task (DeepSWE)
This model has not been measured on DeepSWE (or any other benchmark with a real measured per-task cost) yet.
Artificial Analysis scores (no matched cost)
Artificial Analysis scores below have no measured per-task cost, so no frontier claim is made for them here.
| Benchmark | Score |
|---|---|
| AA intelligence index | 55.3 |
| AA coding index | 76.2 |
| GPQA Diamond (science) | 89.5% |
| Humanity's Last Exam | 37.1% |
| LCR | 68.0% |
| SciCode (coding) | 50.5% |
| Tau2 (banking) | 37.1% |
| Terminal-Bench 2.1 (agentic) | 89.1% |
This page covers 6 reasoning-effort variants of the same model.
| Variant | AA coding index | AA intelligence index | Price /1M |
|---|---|---|---|
| Non-reasoning (shown above) | 76.2 | 55.3 | $20/1M |
| Extra-high effort | 75.9 | 61.0 | $20/1M |
| High effort | 77.1 | 60.3 | $20/1M |
| Low effort | 75.7 | 56.7 | $20/1M |
| Max effort | 76.9 | 61.2 | $20/1M |
| Medium effort | 76.7 | 59.2 | $20/1M |
No LMArena community rating is tracked for this model yet.