GPT-6 Astra
GPT-6 Astra is an API-only (closed weights) language model from OpenAI, released September 3, 2026. The context window is 1.1M tokens. On the Epoch Capabilities Index it scores 169.0, rank 3 of 256 rated models. API price: $10 input / $50 output per million tokens.
Check it against your own PC →
Lab
OpenAI
Released
September 3, 2026
Weights
API-only (closed weights)
Parameters
–
Context
1.1M
Price / M tokens
$10 / $50 in / out
Epoch ECI
169.0 #3
Strengths by category
ModelTape's 0–100 fit on Epoch AI benchmark results.
| Category | Score | Based on |
|---|---|---|
| Reasoning | 55.8 | Chess Puzzles, SimpleBench, Mystery Game Puzzles, Humanity's Last Exam |
| Math | 81.1 | ProofBench, OTIS Mock AIME 2024-2025, FrontierMath-Tiers-1-3-v2-Private, FrontierMath-Tier-4-v2-Private |
| Coding | 79.3 | GSO-Bench, MirrorCode, FrontierSWE, DeepSWE |
| Agent | 40.8 | Balrog, Remote Labor Index |
| Knowledge | 84.3 | GPQA diamond, SimpleQA Verified |
Benchmark results
| Benchmark (Epoch AI) | Score |
|---|---|
| OTIS Mock AIME | 100.0% |
| ProofBench | 99.0% |
| FrontierMath (Tiers 1–3) | 98.6% |
| GPQA Diamond | 98.3% |
| FrontierMath-Tier-4-v2-Private | 97.6% |
| SimpleBench | 83.8% |
| Mystery Game Puzzles | 82.4% |
| GSO-Bench | 79.4% |
| EBR-bench | 76.2% |
| SimpleQA Verified | 75.6% |
| Chess Puzzles | 74.7% |
| DeepSWE | 74.1% |
| Furniture Assembly | 71.4% |
| Balrog | 68.3% |
| FrontierSWE | 65.5% |
| MirrorCode | 60.0% |
| Humanity's Last Exam | 54.8% |
| SpireClimb | 34.0% |
| Remote Labor Index | 20.8% |
Links
Models with a similar score
Gemini 4 Carbon · Claude Opus 5.5 · GPT-6.1 Sol · GPT-6.1 Terra · Claude Fable 5.1 · Claude Sonnet 5.5 · Gemini 4 Argon
More from OpenAI
GPT-6.1 Terra · GPT-6.1 Sol · GPT-6 Sol · GPT-6 Luna · GPT-5.6 Cyber · Daybreak Red · Daybreak Blue · GPT-5.6 Sol