GPT-6 AstravsGPT-5.6 Sol
Across 14 shared benchmarks, GPT-6 Astra leads overall: GPT-6 Astra wins 13, GPT-5.6 Sol wins 1, with 0 ties and an average score difference of +13.21.
GPT-6 Astra
OpenAI · 2026-09-03 · Reasoning model
GPT-5.6 Sol
OpenAI · 2026-06-26 · Reasoning model
GPT-6 Astra13 wins(93%)(7%)1 winGPT-5.6 Sol
Benchmark scores
Grouped by capability, sorted by largest gap within each. 14 shared benchmarks.
General Knowledge
GPT-6 Astra 5/5| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Diff |
|---|---|---|---|
| ARC-AGI-3 | 62.701 / 16Max (No Tools) | 7.803 / 16Max (No Tools) | +54.90 |
| AA Intelligence Index | 61.207 / 28Max (No Tools) | 4224 / 28Normal (With Tools) | +19.20 |
| HLE | 57.2012 / 190Max (With Tools) | 49.5039 / 190Max (No Tools) | +7.70 |
| ARC-AGI-2 | 951 / 85Max (No Tools) | 92.502 / 85Max (No Tools) | +2.50 |
| ARC-AGI-1 | 98.501 / 91Max (No Tools) | 97.504 / 91Extra-High (No Tools) | +1 |
AI Agent - Tool Usage
GPT-6 Astra 3/3| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Diff |
|---|---|---|---|
| Terminal-Bench-Science 0.1 | 64.601 / 11Max (With Tools) | 22.404 / 11Max (With Tools) | +42.20 |
| Terminal-Bench 4.0 | 57.901 / 13Max (With Tools) | 37.276 / 13Max (With Tools) | +20.63 |
| OSWorld 2.0 | 72.602 / 9Max (With Tools) | 62.605 / 9Extra-High (With Tools) | +10 |
Coding and Software Engineer
GPT-6 Astra 2/3| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Diff |
|---|---|---|---|
| AA Coding Agent Index | 674 / 4Max (With Tools) | 801 / 4Extra-High (With Tools) | -13 |
| FrontierCode 1.1 | 64.501 / 6Max (With Tools) | 60.604 / 6Max (With Tools) | +3.90 |
| DeepSWE | 74.102 / 35Max (With Tools) | 72.704 / 35Extra-High (With Tools) | +1.40 |
Agent Level Benchmark
GPT-6 Astra 1/1| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Diff |
|---|---|---|---|
| Agents' Last Exam | 59.301 / 15Max (With Tools) | 52.702 / 15Extra-High (With Tools) | +6.60 |
General Evaluation
GPT-6 Astra 1/1| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Diff |
|---|---|---|---|
| GPQA Diamond | 961 / 271Max (No Tools) | 82.83122 / 271Normal (No Tools) | +13.17 |
Math and Reasoning
GPT-6 Astra 1/1| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Diff |
|---|---|---|---|
| FrontierMath Tier 4 v2 | 97.601 / 41Max (No Tools) | 82.933 / 41Max (No Tools) | +14.67 |
Specs
| Field | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|
| Publisher | OpenAI | OpenAI |
| Release date | 2026-09-03 | 2026-06-26 |
| Model type | Reasoning model | Reasoning model |
| Architecture | Dense | Dense |
| Parameters | Not available | Not available |
| Context length | 1.05M | 1.05M |
| Max output | 128K | 128K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|
| Text input | $10 / 1M tokens | $4 / 1M tokens |
| Text output | $50 / 1M tokens | $20 / 1M tokens |
| Cache read | $1 / 1M tokens | $0.4 / 1M tokens |
| Cache write | $12.5 / 1M tokens | $5 / 1M tokens |
Summary
- GPT-6 Astraleads in:General Knowledge (5/5), AI Agent - Tool Usage (3/3), Coding and Software Engineer (2/3), Agent Level Benchmark (1/1), General Evaluation (1/1), Math and Reasoning (1/1)
On average across the 14 shared benchmarks, GPT-6 Astra scores 13.21 higher.
Largest single-benchmark gap: ARC-AGI-3 — GPT-6 Astra 62.70 vs GPT-5.6 Sol 7.80 (+54.90).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.