GPT-5.6 Sol
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Math95.8
- 2Multimodal#689.8
- 3Instruction following#3189.1
- 4Agentic & tool use#1283.7
- 5Knowledge#1181.6
- 6Reasoning#1376.4
- 7Coding#1164.7
Benchmark results (47)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| Terminal-Bench 2.1 | 91.9 | #2 of 68 |
| BrowseComp | 92.2 | #2 of 50 |
| MazeBench | 13 | #3 of 42 |
| CyberGym | 84.5 | #7 of 25 |
| OSWorld 2.0 | 62.6 | #5 of 24 |
| Toolathlon | 58.0 | #3 of 22 |
| ApprenticeBench | 26 | #5 of 18 |
| ExploitGym | 33.7 | #3 of 14 |
| Terminal-Bench 3.0 | 34.6 | #2 of 13 |
Coding
| SWE-bench Pro | 64.6 | #14 of 69 |
| DeepSWE | 72.7 | #13 of 50 |
| Bug Hunt Bench | 42.0 | #5 of 23 |
| FrontierSWE v2 | 32.2 | #9 of 20 |
| PostTrainBench v1.1 | 36.2 | #6 of 15 |
| VulcanBench v3 | 87.0 | #4 of 14 |
| FrontierCode 1.1 Extended | 60.6 | #5 of 8 |
| VulcanBench CII v1 | 86.5 | #3 of 3 |
Knowledge
| GPQA | 94.6 | #2 of 117 |
| HealthBench Professional | 60.5 | #9 of 16 |
| HealthBench Hard | 33.1 | #5 of 12 |
| LABBench2 | 82.1 | #4 of 7 |
| HLE-Verified | 54.5 | #2 of 6 |
Reasoning
| ARC-AGI-2 | 92.5 | #3 of 41 |
| ARC-AGI-1 | 96.50 | #8 of 36 |
| ARC-AGI-3 | 7.8 | #6 of 22 |
| GeneBench-Pro | 28.7 | #2 of 2 |
Multimodal
| MMMU-Pro | 83.0 | #3 of 48 |
| ParseBench | 75.35 | #3 of 37 |
| ExtractBench (LlamaIndex)(lower is better) | 93.77 | #22 of 24 |
| MMMU-Pro w/ Python | 84.6 | #1 of 9 |
Math
| FrontierMath v2 (Tiers 1-3) | 89.000 | #6 of 64 |
| FrontierMath v2 (Tier 4) | 83.000 | #7 of 51 |
| FrontierMath (legacy) | 89.0 | #1 of 7 |
Instruction following
| Gray Swan IPI (15 attempts)(lower is better) | 27.0 | #9 of 12 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| ✓ Official (OpenAI) checked Oct 17, 2026 · gpt-5.6-sol | $4.00 | $20.00 | $0.40 |
| BenchLM list price | $4.00 | $20.00 | $0.40 |
| OpenRouter openai/gpt-5.6-sol | $4.00 | $20.00 | $0.40 |
Standard tier, prompts under 272K tokens.
OpenAI's API pricing page lists GPT-5.6 Sol at $4.00 input / $0.40 cached input / $20.00 output per million tokens for short-context requests, and states this promotional pricing is available at least through November 21, 2026. Prompts above 272K input tokens use the published $8.00 input / $0.80 cached input / $30.00 output long-context tier; cache writes cost 1.25x uncached input. Observed September 11, 2026; the earlier $5.00 / $30.00 list rate is retained in the price history.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text(xhigh) | 1457 | #43 · 40.4K votes |
| WebDev (coding)(xHigh, Codex Harness) | 1616 | #28 · 17.8K votes |
| Vision(xHigh) | 1281 | #35 · 11.4K votes |
| Search(xhigh) | 1261 | #1 · 60.4K votes |
| Documents(xhigh) | 1483 | #10 · 4.8K votes |
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Same family
- GPT-5.6 Terra70.2
- GPT-5.6 Luna64.4
Similar score
- Muse Wonder Full74.8 · —
- MiMo-V2.6-Pro73.7 · $0.54
- GPT-5.5 Pro73.5 · $67.50
- GPT-6 Sol78.6 · $4.00
- Muse Wonder71.5 · $8.00