Model comparison
GPT-5.6 Luna vs Qwen3.8 Max
Qwen3.8 Max is the stronger model overall, scoring 56.8 to 54.6 on the Noometry Index. GPT-5.6 Luna costs 6.7× less per token, which makes it the better buy when Qwen3.8 Max's lead doesn't matter for your workload.
Last verified . 37 shared benchmarks.
Alibaba (Qwen)Summary
- They share 37 benchmarks with published results for both. GPT-5.6 Luna scores higher in 4 categories and Qwen3.8 Max in 6 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in agentic & tool use, where Qwen3.8 Max leads 45.4 to 34.4.
- The biggest single-benchmark swing is Furniture Assembly: 42.5% for GPT-5.6 Luna and 20% for Qwen3.8 Max.
- GPT-5.6 Luna is cheaper at $0.20 / $1.20 per million input/output tokens, against $2 / $6 for Qwen3.8 Max.
- GPT-5.6 Luna accepts more context: 1.05M tokens versus 1M.
Side by side
| GPT-5.6 Luna | Qwen3.8 Max | |
|---|---|---|
| Provider | OpenAI | Alibaba (Qwen) |
| Noometry Index | 54.6 | 56.8 |
| Released | 2026-07-09 | 2026-08-02 |
| Weights | Proprietary | Proprietary |
| Context window | 1.05M | 1M |
| Max output | 128K | 131K |
| Input $ / M tokens | $0.20 | $2 |
| Output $ / M tokens | $1.20 | $6 |
| Results tracked | 52 | 39 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
GPT-5.6 Luna: 54.5 (#28), Qwen3.8 Max: 53.5 (#29)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| DeepSWE | 67.2% | 57.5% |
| LMArena WebDev | 1519 | 1674 |
| SciCode | 53.6% | 53.2% |
| LMArena Coding | 1466 | 1502 |
| FrontierCode | 39.8% | — |
| CursorBench | 35.9% | — |
| FrontierSWE | — | 17.8% |
| WeirdML | 60.9% | — |
| ALE-Bench | 1,667 | — |
Agentic & Tool Use Qwen3.8 Max leads
GPT-5.6 Luna: 34.4 (#45), Qwen3.8 Max: 45.4 (#14)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| APEX-Agents | 43% | 63.3% |
| GDP.pdf | 22.7% | 23.2% |
| τ²-bench Banking | — | 55.1% |
| BALROG | 45.6% | — |
| Vending-Bench 2 | 4,095 | — |
Reasoning Qwen3.8 Max leads
GPT-5.6 Luna: 47.6 (#43), Qwen3.8 Max: 54.4 (#26)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| NYT Connections (extended) | 69.4% | 88.3% |
| CritPt | 20.6% | 20% |
| Chess Puzzles | 40% | 40% |
| LMArena Hard Prompts | 1451 | 1496 |
| Mystery Game Puzzles | 21% | 38% |
| DTBench | 89.1% | 92% |
| LMCA | 48.5% | 46.2% |
| Epoch Capabilities Index | 156.39 | 156.41 |
| ARC-AGI-2 | 59.5% | — |
| SimpleBench | 46.8% | — |
| Kagi LLM Benchmark | 49.1% | — |
| ARC-AGI-1 | 88% | — |
| Surface Evolver Bench | 61.9% | — |
Math GPT-5.6 Luna leads
GPT-5.6 Luna: 77.7 (#14), Qwen3.8 Max: 73.2 (#20)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| FrontierMath (Tiers 1-3) | 82.1% | 74.7% |
| FrontierMath Tier 4 | 61% | 46.3% |
| OTIS Mock AIME 2024-2025 | 98.3% | 100% |
| ProofBench | 60% | 58% |
| LMArena Math | 1458 | 1499 |
Knowledge Qwen3.8 Max leads
GPT-5.6 Luna: 58.5 (#34), Qwen3.8 Max: 61.7 (#27)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| GPQA Diamond | 91.6% | 92.7% |
| SimpleQA Verified | 41% | 47.3% |
| LMArena Expert | 1478 | 1507 |
Multimodal GPT-5.6 Luna leads
GPT-5.6 Luna: 42.7 (#28), Qwen3.8 Max: 37.2 (#75)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| LMArena Vision | 1258 | 1314 |
| Furniture Assembly | 42.5% | 20% |
| Blueprint-Bench 2 | 22.6% | — |
| LMArena Document | 1457 | — |
Multilingual Qwen3.8 Max leads
GPT-5.6 Luna: 52.8 (#78), Qwen3.8 Max: 56.7 (#18)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| LMArena Non-English | 1417 | 1472 |
| LMArena Chinese | 1470 | 1538 |
| LMArena French | 1456 | 1503 |
| LMArena German | 1454 | 1483 |
| LMArena Japanese | 1411 | 1467 |
| LMArena Korean | 1415 | 1461 |
| LMArena Russian | 1428 | 1481 |
| LMArena Spanish | 1448 | 1492 |
Instruction Following Qwen3.8 Max leads
GPT-5.6 Luna: 75.6 (#57), Qwen3.8 Max: 77.6 (#17)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| LMArena Instruction Following | 1437 | 1479 |
Long Context Qwen3.8 Max leads
GPT-5.6 Luna: 43.9 (#82), Qwen3.8 Max: 45.6 (#31)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| LMArena Longer Query | 1436 | 1489 |
Writing & Preference Too close to call
GPT-5.6 Luna: 68.0 (#29), Qwen3.8 Max: 67.1 (#30)
| Benchmark | GPT-5.6 Luna | Qwen3.8 Max |
|---|---|---|
| LMArena Text | 1431 | 1483 |
| LMArena Creative Writing | 1396 | 1479 |
| LMArena Multi-Turn | 1434 | 1489 |
| EQ-Bench Creative Writing | 1829 | — |
| EQ-Bench 4 | 1156 | — |
Frequently asked questions
Is GPT-5.6 Luna better than Qwen3.8 Max?
Qwen3.8 Max is the stronger model overall, scoring 56.8 to 54.6 on the Noometry Index. GPT-5.6 Luna costs 6.7× less per token, which makes it the better buy when Qwen3.8 Max's lead doesn't matter for your workload.
Which is cheaper, GPT-5.6 Luna or Qwen3.8 Max?
GPT-5.6 Luna is cheaper. It lists at $0.20 per million input tokens and $1.20 per million output tokens; Qwen3.8 Max lists at $2 and $6.
Is GPT-5.6 Luna or Qwen3.8 Max better for coding?
They score almost the same on coding (54.5 vs 53.5); test both on your own repository before choosing.
Which has the bigger context window?
GPT-5.6 Luna does, with 1.05M tokens against 1M.
How many benchmarks do GPT-5.6 Luna and Qwen3.8 Max share?
37 benchmarks have published results for both models. GPT-5.6 Luna has 52 scored results on Noometry and Qwen3.8 Max has 39.