GPT-6 Astra
OpenAI · 2026-09-03
Basic Information
- Developer
- OOpenAI
- Release Date
- 2026-09-03
- Overall Capability Index (ECI)
- 166.5
Official Video
Benchmark Scores (Epoch AI)
Each benchmark also shows the model's rank among the models evaluated.
OPhD-level science questions (GPQA Diamond) · #1 of 30
95.8%
OAdvanced mathematics (FrontierMath) · #1 of 30
93.7%
OExpert-level questions across fields (HLE) · #1 of 30
54.8%
OReasoning through unfamiliar puzzles (ARC-AGI-2) · #1 of 30
95.0%
OFactual question accuracy (SimpleQA Verified) · #1 of 30
75.6%
Bars show accuracy (%) and start at 0%.
User Vote Rankings (LMArena)
Rankings based on people's votes comparing responses from two models. Models do not appear on leaderboards where they have too few votes.
- Overall#29 of 413 · 1477
- English#26 of 413 · 1488
- Web development#2 of 50 · 1788
- Overall agent performance#4 of 50 · 0.123
- Agent task success#6 of 50 · 0.132
- Coding#7 of 408 · 1543
- Document understanding#16 of 44 · 1468
- Image understanding (vision)#20 of 50 · 1284
- French#26 of 290 · 1497
- Math#27 of 396 · 1490
- Creative writing#39 of 411 · 1449
- Chinese#43 of 389 · 1510
- Japanese#68 of 274 · 1418
- Spanish#86 of 295 · 1437
Where to Use It
Models from OpenAI are primarily available through ChatGPT. Available models and usage limits vary by plan, so check the pricing and free access page.
Scores and rankings are reproduced directly from public sources and were not measured by this site. Model names vary slightly across sources and are matched automatically, so different versions may occasionally be mixed together.
Source: Epoch AI (CC BY 4.0) · Checked 2026-10-04 · LMArena (CC BY 4.0) · Checked 2026-10-04