GPT-5.6 Luna
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Math84.0
- 2Multimodal#1979.6
- 3Knowledge70.9
- 4Agentic & tool use70.2
- 5Reasoning#2658.2
- 6Coding#1656.4
Benchmark results (29)
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 | 84.7 | #18 of 68 |
| BrowseComp | 83.3 | #19 of 50 |
| MazeBench | 0 | #20 of 42 |
| CyberGym | 77.9 | #16 of 25 |
| OSWorld 2.0 | 45.6 | #12 of 24 |
| Toolathlon | 53.4 | #7 of 22 |
| ApprenticeBench | 7 | #15 of 18 |
| ExploitGym | 12.4 | #10 of 14 |
| Terminal-Bench 3.0 | 14.3 | #10 of 13 |
Coding
| SWE-bench Pro | 62.7 | #19 of 69 |
| DeepSWE | 67.2 | #26 of 50 |
| VulcanBench v3 | 85.5 | #9 of 14 |
| FrontierCode 1.1 Extended | 55.1 | #8 of 8 |
Multimodal
| MMMU-Pro | 78.4 | #20 of 48 |
| ParseBench | 68.34 | #18 of 37 |
| ExtractBench (LlamaIndex)(lower is better) | 91.16 | #18 of 24 |
| MMMU-Pro w/ Python | 79.5 | #7 of 9 |
Reasoning
| ARC-AGI-2 | 59.5 | #29 of 41 |
| ARC-AGI-1 | 88.00 | #25 of 36 |
| ARC-AGI-3 | 0.2 | #18 of 22 |
Knowledge
| GPQA | 92.3 | #14 of 117 |
| HealthBench Professional | 55.7 | #15 of 16 |
| HealthBench Hard | 32.0 | #7 of 12 |
Math
| FrontierMath v2 (Tiers 1-3) | 78.600 | #17 of 64 |
| FrontierMath v2 (Tier 4) | 58.500 | #15 of 51 |
| FrontierMath (legacy) | 78.6 | #3 of 7 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| ✓ Official (OpenAI) checked Oct 17, 2026 · gpt-5.6-luna | $0.20 | $1.20 | $0.020 |
| BenchLM list price | $0.20 | $1.20 | $0.020 |
| OpenRouter openai/gpt-5.6-luna | $0.20 | $1.20 | $0.020 |
Standard tier, prompts under 272K tokens.
OpenAI's API pricing page lists GPT-5.6 Luna at $0.20 input / $0.02 cached input / $1.20 output per million tokens for short-context requests. Prompts above 272K input tokens use the published $0.40 input / $0.04 cached input / $1.80 output long-context tier; cache writes cost $0.25 per million tokens. Observed September 11, 2026; the $1.00 / $6.00 launch rate is retained in the price history.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text(xhigh) | 1431 | #93 · 41.1K votes |
| WebDev (coding)(xHigh, Codex Harness) | 1518 | #56 · 12.5K votes |
| Vision(xHigh) | 1257 | #58 · 11.1K votes |
| Documents(xhigh) | 1457 | #23 · 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 Terra70.2
Similar score
- MiMo-V2.6-Flash64.6 · $0.17
- Gemini 3.1 Pro64.7 · $4.50
- dots3-note Preview64.1 · —
- Claude Sonnet 564.9 · $4.00
- Claude Opus 4.765.5 · $10.00