Model comparison
GPT-5.4 mini vs Qwen3.8 Max
Qwen3.8 Max is the stronger model overall, scoring 56.8 to 45.0 on the Noometry Index. GPT-5.4 mini costs 1.8× less per token, which makes it the better buy when Qwen3.8 Max's lead doesn't matter for your workload.
Last verified . 33 shared benchmarks.
Alibaba (Qwen)Summary
- They share 33 benchmarks with published results for both. GPT-5.4 mini scores higher in 1 category and Qwen3.8 Max in 9 categories; 10 gaps are clear of the uncertainty.
- The widest gap is in math, where Qwen3.8 Max leads 73.2 to 45.5.
- The biggest single-benchmark swing is ProofBench: 21% for GPT-5.4 mini and 58% for Qwen3.8 Max.
- GPT-5.4 mini is cheaper at $0.75 / $4.50 per million input/output tokens, against $2 / $6 for Qwen3.8 Max.
- Qwen3.8 Max accepts more context: 1M tokens versus 400K.
Side by side
| GPT-5.4 mini | Qwen3.8 Max | |
|---|---|---|
| Provider | OpenAI | Alibaba (Qwen) |
| Noometry Index | 45.0 | 56.8 |
| Released | 2026-03-17 | 2026-08-02 |
| Weights | Proprietary | Proprietary |
| Context window | 400K | 1M |
| Max output | 128K | 131K |
| Input $ / M tokens | $0.75 | $2 |
| Output $ / M tokens | $4.50 | $6 |
| Results tracked | 46 | 39 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Qwen3.8 Max leads
GPT-5.4 mini: 45.2 (#72), Qwen3.8 Max: 53.5 (#29)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| LMArena WebDev | 1397 | 1674 |
| SciCode | 49.9% | 53.2% |
| LMArena Coding | 1438 | 1502 |
| DeepSWE | — | 57.5% |
| FrontierCode | 27% | — |
| FrontierSWE | — | 17.8% |
| WeirdML | 60.3% | — |
| ALE-Bench | 1,189 | — |
Agentic & Tool Use Qwen3.8 Max leads
GPT-5.4 mini: 29.9 (#81), Qwen3.8 Max: 45.4 (#14)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| APEX-Agents | — | 63.3% |
| τ²-bench Banking | — | 55.1% |
| DeepResearch Bench | 36.3% | — |
| GDP.pdf | — | 23.2% |
Reasoning Qwen3.8 Max leads
GPT-5.4 mini: 30.4 (#85), Qwen3.8 Max: 54.4 (#26)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| NYT Connections (extended) | 61.8% | 88.3% |
| CritPt | 10% | 20% |
| Chess Puzzles | 24% | 40% |
| LMArena Hard Prompts | 1424 | 1496 |
| Mystery Game Puzzles | 11% | 38% |
| DTBench | 80% | 92% |
| LMCA | 40.8% | 46.2% |
| Epoch Capabilities Index | 148.84 | 156.41 |
| ARC-AGI-2 | 18.9% | — |
| Kagi LLM Benchmark | 37.9% | — |
| ARC-AGI-1 | 63.7% | — |
| Thematic Generalization | 61.7% | — |
| ForecastBench | 57 | — |
Math Qwen3.8 Max leads
GPT-5.4 mini: 45.5 (#75), Qwen3.8 Max: 73.2 (#20)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| FrontierMath (Tiers 1-3) | 51.2% | 74.7% |
| FrontierMath Tier 4 | 9.8% | 46.3% |
| OTIS Mock AIME 2024-2025 | 88.9% | 100% |
| ProofBench | 21% | 58% |
| LMArena Math | 1419 | 1499 |
| FrontierMath (Feb 2025 set) | 28.3% | — |
| FrontierMath Tier 4 (v1) | 2.1% | — |
Knowledge Qwen3.8 Max leads
GPT-5.4 mini: 51.5 (#67), Qwen3.8 Max: 61.7 (#27)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| GPQA Diamond | 86.9% | 92.7% |
| SimpleQA Verified | 29.4% | 47.3% |
| LMArena Expert | 1435 | 1507 |
| Vectara Hallucination Rate | 5.5% | — |
Multimodal GPT-5.4 mini leads
GPT-5.4 mini: 39.7 (#56), Qwen3.8 Max: 37.2 (#75)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| LMArena Vision | 1245 | 1314 |
| Furniture Assembly | — | 20% |
Multilingual Qwen3.8 Max leads
GPT-5.4 mini: 51.9 (#96), Qwen3.8 Max: 56.7 (#18)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| LMArena Non-English | 1405 | 1472 |
| LMArena Chinese | 1446 | 1538 |
| LMArena French | 1440 | 1503 |
| LMArena German | 1409 | 1483 |
| LMArena Japanese | 1374 | 1467 |
| LMArena Korean | 1368 | 1461 |
| LMArena Russian | 1417 | 1481 |
| LMArena Spanish | 1405 | 1492 |
Instruction Following Qwen3.8 Max leads
GPT-5.4 mini: 74.1 (#102), Qwen3.8 Max: 77.6 (#17)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| LMArena Instruction Following | 1405 | 1479 |
Long Context Qwen3.8 Max leads
GPT-5.4 mini: 43.0 (#112), Qwen3.8 Max: 45.6 (#31)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| LMArena Longer Query | 1407 | 1489 |
Writing & Preference Qwen3.8 Max leads
GPT-5.4 mini: 64.0 (#58), Qwen3.8 Max: 67.1 (#30)
| Benchmark | GPT-5.4 mini | Qwen3.8 Max |
|---|---|---|
| LMArena Text | 1412 | 1483 |
| LMArena Creative Writing | 1370 | 1479 |
| LMArena Multi-Turn | 1429 | 1489 |
| EQ-Bench Creative Writing | 1665 | — |
Frequently asked questions
Is GPT-5.4 mini better than Qwen3.8 Max?
Qwen3.8 Max is the stronger model overall, scoring 56.8 to 45.0 on the Noometry Index. GPT-5.4 mini costs 1.8× 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-5.4 mini or Qwen3.8 Max?
GPT-5.4 mini is cheaper. It lists at $0.75 per million input tokens and $4.50 per million output tokens; Qwen3.8 Max lists at $2 and $6.
Is GPT-5.4 mini or Qwen3.8 Max better for coding?
Qwen3.8 Max scores higher on coding benchmarks: 53.5 versus 45.2 in the Noometry coding category.
Which has the bigger context window?
Qwen3.8 Max does, with 1M tokens against 400K.
How many benchmarks do GPT-5.4 mini and Qwen3.8 Max share?
33 benchmarks have published results for both models. GPT-5.4 mini has 46 scored results on Noometry and Qwen3.8 Max has 39.