Model comparison
Qwen3.8 Max vs Step 5 Preview
Qwen3.8 Max is the stronger model overall, scoring 56.8 to 47.9 on the Noometry Index. Step 5 Preview costs 2.1× less per token, which makes it the better buy when Qwen3.8 Max's lead doesn't matter for your workload.
Last verified . 18 shared benchmarks.
Alibaba (Qwen)
StepFunSummary
- They share 18 benchmarks with published results for both. Qwen3.8 Max scores higher in 8 categories and Step 5 Preview in 1 category; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where Qwen3.8 Max leads 73.2 to 46.1.
- The biggest single-benchmark swing is ProofBench: 58% for Qwen3.8 Max and 42% for Step 5 Preview.
- Step 5 Preview is cheaper at $1 / $2.70 per million input/output tokens, against $2 / $6 for Qwen3.8 Max.
- Step 5 Preview accepts more context: 1.02M tokens versus 1M.
Side by side
| Qwen3.8 Max | Step 5 Preview | |
|---|---|---|
| Provider | Alibaba (Qwen) | StepFun |
| Noometry Index | 56.8 | 47.9 |
| Released | 2026-08-02 | 2026-09-16 |
| Weights | Proprietary | Proprietary |
| Context window | 1M | 1.02M |
| Max output | 131K | 66K |
| Input $ / M tokens | $2 | $1 |
| Output $ / M tokens | $6 | $2.70 |
| Results tracked | 39 | 18 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Qwen3.8 Max leads
Qwen3.8 Max: 53.5 (#29), Step 5 Preview: 51.8 (#36)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| LMArena WebDev | 1674 | 1564 |
| SciCode | 53.2% | 58.9% |
| LMArena Coding | 1502 | 1480 |
| DeepSWE | 57.5% | — |
| FrontierSWE | 17.8% | — |
Agentic & Tool Use Not comparable
Qwen3.8 Max: 45.4 (#14), Step 5 Preview: —
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| APEX-Agents | 63.3% | — |
| τ²-bench Banking | 55.1% | — |
| GDP.pdf | 23.2% | — |
Reasoning Qwen3.8 Max leads
Qwen3.8 Max: 54.4 (#26), Step 5 Preview: 40.0 (#57)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| CritPt | 20% | 20.9% |
| LMArena Hard Prompts | 1496 | 1465 |
| NYT Connections (extended) | 88.3% | — |
| Chess Puzzles | 40% | — |
| Mystery Game Puzzles | 38% | — |
| DTBench | 92% | — |
| LMCA | 46.2% | — |
| Epoch Capabilities Index | 156.41 | — |
Math Qwen3.8 Max leads
Qwen3.8 Max: 73.2 (#20), Step 5 Preview: 46.1 (#72)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| ProofBench | 58% | 42% |
| LMArena Math | 1499 | 1470 |
| FrontierMath (Tiers 1-3) | 74.7% | — |
| FrontierMath Tier 4 | 46.3% | — |
| OTIS Mock AIME 2024-2025 | 100% | — |
Knowledge Qwen3.8 Max leads
Qwen3.8 Max: 61.7 (#27), Step 5 Preview: 41.2 (#112)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| LMArena Expert | 1507 | 1470 |
| GPQA Diamond | 92.7% | — |
| SimpleQA Verified | 47.3% | — |
Multimodal Step 5 Preview leads
Qwen3.8 Max: 37.2 (#75), Step 5 Preview: 41.0 (#41)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| LMArena Vision | 1314 | 1267 |
| Furniture Assembly | 20% | — |
Multilingual Qwen3.8 Max leads
Qwen3.8 Max: 56.7 (#18), Step 5 Preview: 53.8 (#54)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| LMArena Non-English | 1472 | 1432 |
| LMArena Chinese | 1538 | 1519 |
| LMArena Russian | 1481 | 1436 |
| LMArena Spanish | 1492 | 1439 |
| LMArena French | 1503 | — |
| LMArena German | 1483 | — |
| LMArena Japanese | 1467 | — |
| LMArena Korean | 1461 | — |
Instruction Following Qwen3.8 Max leads
Qwen3.8 Max: 77.6 (#17), Step 5 Preview: 76.0 (#50)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| LMArena Instruction Following | 1479 | 1444 |
Long Context Qwen3.8 Max leads
Qwen3.8 Max: 45.6 (#31), Step 5 Preview: 44.5 (#64)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| LMArena Longer Query | 1489 | 1455 |
Writing & Preference Qwen3.8 Max leads
Qwen3.8 Max: 67.1 (#30), Step 5 Preview: 62.8 (#71)
| Benchmark | Qwen3.8 Max | Step 5 Preview |
|---|---|---|
| LMArena Text | 1483 | 1442 |
| LMArena Creative Writing | 1479 | 1410 |
| LMArena Multi-Turn | 1489 | 1447 |
Frequently asked questions
Is Qwen3.8 Max better than Step 5 Preview?
Qwen3.8 Max is the stronger model overall, scoring 56.8 to 47.9 on the Noometry Index. Step 5 Preview costs 2.1× less per token, which makes it the better buy when Qwen3.8 Max's lead doesn't matter for your workload.
Which is cheaper, Qwen3.8 Max or Step 5 Preview?
Step 5 Preview is cheaper. It lists at $1 per million input tokens and $2.70 per million output tokens; Qwen3.8 Max lists at $2 and $6.
Is Qwen3.8 Max or Step 5 Preview better for coding?
Qwen3.8 Max scores higher on coding benchmarks: 53.5 versus 51.8 in the Noometry coding category.
Which has the bigger context window?
Step 5 Preview does, with 1.02M tokens against 1M.
How many benchmarks do Qwen3.8 Max and Step 5 Preview share?
18 benchmarks have published results for both models. Qwen3.8 Max has 39 scored results on Noometry and Step 5 Preview has 18.