Model comparison
GPT-5.4 vs Qwen3.8 Max
GPT-5.4 is the stronger model overall, scoring 59.5 to 56.0 on the Noometry Index. Qwen3.8 Max costs 1.9× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 36 shared benchmarks.
Alibaba (Qwen)Summary
- They share 36 benchmarks with published results for both. GPT-5.4 scores higher in 8 categories and Qwen3.8 Max in 2 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.4 leads 61.7 to 53.4.
- The biggest single-benchmark swing is Furniture Assembly: 37.5% for GPT-5.4 and 20% for Qwen3.8 Max.
- Qwen3.8 Max is cheaper at $2 / $6 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 1M.
Side by side
| GPT-5.4 | Qwen3.8 Max | |
|---|---|---|
| Provider | OpenAI | Alibaba (Qwen) |
| Noometry Index | 59.5 | 56.0 |
| Released | 2026-03-05 | 2026-08-02 |
| Weights | Proprietary | Proprietary |
| Context window | 1.05M | 1M |
| Max output | 128K | 131K |
| Input $ / M tokens | $2.50 | $2 |
| Output $ / M tokens | $15 | $6 |
| Results tracked | 68 | 38 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
GPT-5.4: 52.8 (#28), Qwen3.8 Max: 52.3 (#30)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| DeepSWE | 51.8% | 57.5% |
| LMArena WebDev | 1465 | 1671 |
| SciCode | 56.6% | 52.9% |
| LMArena Coding | 1497 | 1499 |
| SWE-bench Verified | 76.9% | — |
| FrontierSWE | — | 15.8% |
| GSO | 31.4% | — |
| WeirdML | 77.7% | — |
| MirrorCode | 15.6% | — |
| ALE-Bench | 1,607 | — |
| AlgoTune | 1.85 | — |
Agentic & Tool Use GPT-5.4 leads
GPT-5.4: 46.5 (#12), Qwen3.8 Max: 42.6 (#17)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| τ²-bench Banking | 39.4% | 55.1% |
| Terminal-Bench | 81.8% | — |
| APEX-Agents | 52.4% | — |
| DeepResearch Bench | 35.1% | — |
| PostTrainBench | 19% | — |
| GBAEval | 45.1% | — |
| GDP.pdf | — | 23.2% |
| LMArena Search | 1197 | — |
| METR Time Horizons | 74.3% | — |
| Vending-Bench 2 | 6,144 | — |
Reasoning GPT-5.4 leads
GPT-5.4: 61.7 (#17), Qwen3.8 Max: 53.4 (#26)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| NYT Connections (extended) | 91.3% | 88.3% |
| CritPt | 23.4% | 20% |
| Chess Puzzles | 44% | 40% |
| LMArena Hard Prompts | 1485 | 1493 |
| Mystery Game Puzzles | 37% | 38% |
| DTBench | 94.4% | 92% |
| LMCA | 52% | 46.2% |
| Epoch Capabilities Index | 156.92 | 156.69 |
| ARC-AGI-2 | 74% | — |
| Kagi LLM Benchmark | 63.8% | — |
| ARC-AGI-1 | 93.7% | — |
| EnigmaEval | 16% | — |
| Thematic Generalization | 80% | — |
| EBR-Bench | 25.4% | — |
| ForecastBench | 59.5 | — |
Math Too close to call
GPT-5.4: 73.3 (#14), Qwen3.8 Max: 73.1 (#15)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| FrontierMath (Tiers 1-3) | 78.6% | 74.7% |
| FrontierMath Tier 4 | 49% | 46.3% |
| OTIS Mock AIME 2024-2025 | 97.8% | 100% |
| ProofBench | 56% | 58% |
| LMArena Math | 1488 | 1498 |
| MathArena Final-Answer Competitions | 83.1% | — |
| FrontierMath (Feb 2025 set) | 47.6% | — |
| FrontierMath Tier 4 (v1) | 27.1% | — |
Knowledge GPT-5.4 leads
GPT-5.4: 65.3 (#13), Qwen3.8 Max: 61.7 (#25)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| GPQA Diamond | 93.3% | 92.7% |
| SimpleQA Verified | 45.1% | 47.3% |
| LMArena Expert | 1507 | 1504 |
| Humanity's Last Exam | 36.2% | — |
| Vectara Hallucination Rate | 7% | — |
Multimodal GPT-5.4 leads
GPT-5.4: 43.8 (#19), Qwen3.8 Max: 36.9 (#70)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| LMArena Vision | 1302 | 1314 |
| Furniture Assembly | 37.5% | 20% |
| Blueprint-Bench 2 | 27.1% | — |
| LMArena Document | 1471 | — |
Multilingual Too close to call
GPT-5.4: 56.4 (#22), Qwen3.8 Max: 56.6 (#18)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| LMArena Non-English | 1465 | 1468 |
| LMArena Chinese | 1520 | 1537 |
| LMArena French | 1493 | 1502 |
| LMArena German | 1472 | 1489 |
| LMArena Japanese | 1484 | 1466 |
| LMArena Korean | 1449 | 1467 |
| LMArena Russian | 1480 | 1478 |
| LMArena Spanish | 1453 | 1495 |
Instruction Following Too close to call
GPT-5.4: 77.3 (#25), Qwen3.8 Max: 77.7 (#19)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| LMArena Instruction Following | 1469 | 1478 |
Long Context GPT-5.4 leads
GPT-5.4: 50.3 (#8), Qwen3.8 Max: 45.5 (#30)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| LMArena Longer Query | 1474 | 1486 |
| CL-bench | 27.9% | — |
| CL-bench Life | 21.7% | — |
Writing & Preference GPT-5.4 leads
GPT-5.4: 72.1 (#18), Qwen3.8 Max: 67.0 (#31)
| Benchmark | GPT-5.4 | Qwen3.8 Max |
|---|---|---|
| LMArena Text | 1469 | 1480 |
| LMArena Creative Writing | 1439 | 1478 |
| LMArena Multi-Turn | 1482 | 1487 |
| EQ-Bench Creative Writing | 1840 | — |
| EQ-Bench 4 | 1272 | — |
Frequently asked questions
Is GPT-5.4 better than Qwen3.8 Max?
GPT-5.4 is the stronger model overall, scoring 59.5 to 56.0 on the Noometry Index. Qwen3.8 Max costs 1.9× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Which is cheaper, GPT-5.4 or Qwen3.8 Max?
Qwen3.8 Max is cheaper. It lists at $2 per million input tokens and $6 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is GPT-5.4 or Qwen3.8 Max better for coding?
They score almost the same on coding (52.8 vs 52.3); test both on your own repository before choosing.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 1M.
How many benchmarks do GPT-5.4 and Qwen3.8 Max share?
36 benchmarks have published results for both models. GPT-5.4 has 68 scored results on Noometry and Qwen3.8 Max has 38.