Qwen3.8 Max (0902) vs Qwen3 Max Thinking (Preview)
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
Qwen Max (0902)43.1
Qwen3 Max Thinking (Preview)12.7
Terminal-Bench 4.0 Terminal-Bench 4.0 · source setting 2: Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite
Qwen Max (0902)38.9
Qwen3 Max Thinking (Preview)—
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
Qwen Max (0902)56.2
Qwen3 Max Thinking (Preview)—
MMMU-Pro (vision) MMMU-Pro (vision): College-level questions that require understanding images: charts, diagrams, photos
Qwen Max (0902)82.8
Qwen3 Max Thinking (Preview)—
| Benchmark | Qwen Max (0902) | Qwen3 Max Thinking (Preview) |
|---|---|---|
| Humanity's Last Exam | 43.1 | 12.7 |
| Terminal-Bench 4.0 | 38.9 | — |
| AutomationBench-AA | 56.2 | — |
| MMMU-Pro (vision) | 82.8 | — |
Specs
| Spec | Qwen Max (0902) | Qwen3 Max Thinking (Preview) |
|---|---|---|
| Input, per 1M | !$2 | — |
| Output, per 1M | !$6 | — |
| Cost per task? | $5.41 | — |
| Speed, tok/s | 40 | — |
| Context | 984K | 262K |
| Weights | Closed | Closed |
| Released | Sep 2026 | Nov 2025 |
| See images | ✓ | — |
| Read PDFs | ? | ? |
| Hear audio | — | — |
| Watch video | ✓ | — |
| Call tools | ✓ | ? |
| Force a tool call | ? | ? |
| Return strict JSON | ✓ | ? |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
Qwen Max (0902) → Qwen3 Max Thinking (Preview)
Context: 984K → 262K
No image input
Qwen3 Max Thinking (Preview) → Qwen Max (0902)
Nothing known.