Qwen 3.5 9B vs Qwen3 VL 8B Instruct
Different efforts: thinking · non-thinking
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
Qwen 3.5 9B14.9
Qwen3 VL 8B Instruct2.7
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 3.5 9B0.5
Qwen3 VL 8B Instruct—
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
Qwen 3.5 9B1
Qwen3 VL 8B Instruct—
MMMU-Pro (vision) MMMU-Pro (vision): College-level questions that require understanding images: charts, diagrams, photos
Qwen 3.5 9B69.2
Qwen3 VL 8B Instruct47.3
| Benchmark | Qwen 3.5 9B | Qwen3 VL 8B Instruct |
|---|---|---|
| Humanity's Last Exam | 14.9 | 2.7 |
| Terminal-Bench 4.0 | 0.5 | — |
| AutomationBench-AA | 1 | — |
| MMMU-Pro (vision) | 69.2 | 47.3 |
Specs
| Spec | Qwen 3.5 9B | Qwen3 VL 8B Instruct |
|---|---|---|
| Input, per 1M | $0.08 | $0.18 |
| Output, per 1M | $0.13 | $0.69 |
| Cost per task? | $0.21 | — |
| Speed, tok/s | 72 | — |
| Context | 262K | 256K |
| Weights | Open | Open |
| Released | Mar 2026 | Oct 2025 |
| Memory (4-bit) | 4.8 GB | 4.4 GB |
| Licence | Apache-2.0 | Apache-2.0 |
| Commercial use | Allowed | Allowed |
| See images | ✓ | ✓ |
| Read PDFs | — | — |
| Hear audio | — | — |
| Watch video | ✓ | — |
| Call tools | ✓ | ✓ |
| Force a tool call | Partial | Partial |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
Qwen 3.5 9B → Qwen3 VL 8B Instruct
Context: 262K → 256K
Qwen3 VL 8B Instruct → Qwen 3.5 9B