Compare / MiniCPM5 2B vs Ministral 3 3B
MiniCPM5 2B vs Ministral 3 3B
Updated November 2026
Verdict
MiniCPM5 2B scores higher on the Small LM Index: 57.5 vs 49.4. On the 3 benchmarks both labs report, MiniCPM5 2B wins 3 and Ministral 3 3B wins 0. Both need about 3.3 GB of memory at 4-bit quantization. Only Ministral 3 3B has a hosted API on OpenRouter ($0.10 per million tokens); the other has to be self-hosted.
Pick MiniCPM5 2B if…
- you want the strongest model overall
- you mostly write code
- you need math and reasoning
- you want the more widely used model (more quantized builds and community support)
Pick Ministral 3 3B if…
- you want a hosted API rather than running it yourself
- you work with long documents
- you need image input
Specifications
| MiniCPM5 2B | Ministral 3 3B | |
|---|---|---|
| Small LM Index | 57.5 (likely 56–59) | 49.4 (likely 45–54) |
| Lab | OpenBMB (MiniCPM) · China | Mistral AI · Europe |
| Released | 2026-09-06 | 2025-12-02 |
| Parameters | 2.5B (dense) | 3B (dense) |
| Memory at 4-bit | ~2.4 GB | ~3.3 GB |
| Context window | 131K tokens | 256K tokens |
| Input | text | text, image |
| Reasoning | Hybrid (thinking on/off) | Hybrid (thinking on/off) |
| License | Apache-2.0 | Apache-2.0 |
| API price / 1M tokens | Self-host only | $0.10 ($0.10 in, $0.10 out) |
| Hugging Face downloads (30 days) | 1.1M | 17.7K |
Benchmarks
Higher is better. Bold marks the winner where both labs report a score. ≈ marks the index model's estimate for a score that wasn't reported, ✓ an independent measurement and * a score measured by another lab.
| Benchmark | MiniCPM5 2B | Ministral 3 3B |
|---|---|---|
| MMLU-Pro Knowledge | 70.8 | ≈ 72 |
| GPQA Diamond Knowledge | 70.2 | 53.4 |
| Humanity's Last Exam Reasoning | 8.9 | ≈ 6 |
| AIME Math | 86.5 | 72.1 |
| LiveCodeBench Coding | 69.1 | 54.8 |
| SWE-bench Verified Coding | 46.4 | ≈ 30 |
| Terminal-Bench Agentic | 8.6 | — |
| τ²-Bench Agentic | 97.1 | — |
| IFBench Instruction following | 66.3 | — |
| MMLU Knowledge | 84.7 | — |
| IFEval Instruction following | 86.7 | — |
| BFCL Agentic | 66.6 | — |
| MATH-500 Math | 94.6 | — |
| HMMT Math | 63.8 | — |
| AA-LCR Long context | 59.0 | — |
Scores are mostly reported by each lab with its own settings, so small gaps are noise. See the methodology. Open in the interactive comparison to add more models.
Questions
Which is better, MiniCPM5 2B or Ministral 3 3B?
MiniCPM5 2B scores higher on the Small LM Index: 57.5 vs 49.4. On the 3 benchmarks both labs report, MiniCPM5 2B wins 3 and Ministral 3 3B wins 0.
Which is cheaper to run?
Only Ministral 3 3B has a hosted API on OpenRouter ($0.10 per million tokens); the other has to be self-hosted. Both need about 3.3 GB of memory at 4-bit quantization.
Which has the longer context window?
Ministral 3 3B has the longer context window: 256K tokens against 131K.
Can I use MiniCPM5 2B and Ministral 3 3B commercially?
Both are released under the Apache-2.0 license. Apache 2.0 and MIT allow commercial use; other licenses can add conditions, so read the license before shipping a product.