Devstral Small 2 vs Mistral Medium 3.5
Different efforts: non-thinking · high
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
Devstral Small 23.5
Mistral Medium 3.513.8
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
Devstral Small 20
Mistral Medium 3.50
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
Devstral Small 23.1
Mistral Medium 3.56.3
Arena Text · creative writing Arena Text · creative writing: Human pairwise preferences in the stated text vote pool
Devstral Small 2—
Mistral Medium 3.51,400
MMMU-Pro (vision) MMMU-Pro (vision): College-level questions that require understanding images: charts, diagrams, photos
Devstral Small 244.6
Mistral Medium 3.564.9
| Benchmark | Devstral Small 2 | Mistral Medium 3.5 |
|---|---|---|
| Humanity's Last Exam | 3.5 | 13.8 |
| Terminal-Bench 4.0 | 0 | 0 |
| AutomationBench-AA | 3.1 | 6.3 |
| Arena Text · creative writing | — | 1,400 |
| MMMU-Pro (vision) | 44.6 | 64.9 |
Specs
| Spec | Devstral Small 2 | Mistral Medium 3.5 |
|---|---|---|
| Input, per 1M | !$0 | $1.50 |
| Output, per 1M | !$0 | $7.50 |
| Cost per task? | — | $0.50 |
| Speed, tok/s | — | 155 |
| Context | 256K | 262K |
| Weights | Open | Open |
| Released | Dec 2025 | Apr 2026 |
| Memory (4-bit) | 14 GB | 77 GB |
| Licence | Apache-2.0 | custom licence |
| Commercial use | Allowed | Restricted |
| See images | ✓ | ✓ |
| Read PDFs | — | ✓ |
| Hear audio | — | — |
| Watch video | — | — |
| Call tools | ✓ | ✓ |
| Force a tool call | ? | Partial |
| Return strict JSON | Partial | ✓ |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
Devstral Small 2 → Mistral Medium 3.5
Memory (4-bit): 14 GB → 77 GB
Mistral Medium 3.5 → Devstral Small 2