GPT-5.6 Terra
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Math90.1
- 2Instruction following#3987.1
- 3Multimodal#1184.5
- 4Agentic & tool use#2776.0
- 5Knowledge#1873.7
- 6Reasoning#1969.1
- 7Coding#1564.3
Benchmark results (31)
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 | 87.4 | #12 of 68 |
| BrowseComp | 87.5 | #11 of 50 |
| MazeBench | 6 | #11 of 42 |
| CyberGym | 81.8 | #12 of 25 |
| OSWorld 2.0 | 50.2 | #9 of 24 |
| Toolathlon | 53.1 | #8 of 22 |
| ApprenticeBench | 16 | #10 of 18 |
| ExploitGym | 23.2 | #4 of 14 |
| Terminal-Bench 3.0 | 20.8 | #6 of 13 |
Knowledge
| GPQA | 92.9 | #8 of 117 |
| HealthBench Professional | 57.7 | #13 of 16 |
| HealthBench Hard | 32.7 | #6 of 12 |
| LABBench2 | 81.2 | #6 of 7 |
| HLE-Verified | 51.1 | #5 of 6 |
Coding
| SWE-bench Pro | 63.4 | #16 of 69 |
| DeepSWE | 69.6 | #19 of 50 |
| VulcanBench v3 | 87.0 | #4 of 14 |
| FrontierCode 1.1 Extended | 55.8 | #7 of 8 |
Multimodal
| MMMU-Pro | 80.7 | #10 of 48 |
| ParseBench | 68.39 | #17 of 37 |
| ExtractBench (LlamaIndex)(lower is better) | 92.26 | #20 of 24 |
| MMMU-Pro w/ Python | 82.0 | #5 of 9 |
Reasoning
| ARC-AGI-2 | 83.9 | #14 of 41 |
| ARC-AGI-1 | 96.50 | #8 of 36 |
| ARC-AGI-3 | 0.8 | #11 of 22 |
Math
| FrontierMath v2 (Tiers 1-3) | 84.900 | #13 of 64 |
| FrontierMath v2 (Tier 4) | 68.300 | #13 of 51 |
| FrontierMath (legacy) | 84.9 | #2 of 7 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| ✓ Official (OpenAI) checked Oct 17, 2026 · gpt-5.6-terra | $2.00 | $12.00 | $0.20 |
| BenchLM list price | $2.00 | $12.00 | $0.20 |
| OpenRouter openai/gpt-5.6-terra | $2.00 | $12.00 | $0.20 |
Standard tier, prompts under 272K tokens.
OpenAI's API pricing page lists GPT-5.6 Terra at $2.00 input / $0.20 cached input / $12.00 output per million tokens for short-context requests. Prompts above 272K input tokens use the published $4.00 input / $0.40 cached input / $18.00 output long-context tier; cache writes cost $2.50 per million tokens. Observed September 11, 2026; the $2.50 / $15.00 launch rate is retained in the price history.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text(xhigh) | 1447 | #58 · 43.3K votes |
| WebDev (coding)(xHigh, Codex Harness) | 1522 | #54 · 13.9K votes |
| Vision(xHigh) | 1271 | #43 · 11.8K votes |
| Documents(xhigh) | 1472 | #13 · 5.8K votes |
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Same family
- GPT-5.6 Sol75.3
- GPT-5.6 Luna64.4
Similar score
- Claude Opus 4.870.0 · $10.00
- Kimi K370.0 · $6.00
- Qwen3.8 Max69.7 · —
- Step 5 Preview69.6 · $1.43
- GPT-5.569.2 · $11.25