Model Details
MiniCPM SALA
Quick answer: MiniCPM SALA is OpenBMB's July 2025 model scoring 95.1% HumanEval and 67% MMLU-Pro. Apache 2.0 — exceptional HumanEval coding.
At a Glance
Where MiniCPM SALA leads
- 95.1% HumanEval — top-tier coding benchmark
- Apache 2.0 — fully open
- OpenBMB's compact efficiency focus
Where it lags
- 67% MMLU-Pro — moderate academic breadth
- Smaller community than Qwen/Mistral
Best for: HumanEval-optimised open coding tasks; Apache 2.0 maximum-HumanEval compact model.
What MiniCPM SALA Is
MiniCPM SALA is part of OpenBMB's compact model line, July 2025. The 95.1% HumanEval is outstanding — surpassing many larger models. SALA likely stands for a specialized alignment/coding technique from OpenBMB.
Specifications
| Field | Value |
|---|---|
| Organization | OpenBMB |
| License | Apache 2.0 |
| HuggingFace | openbmb/MiniCPM-SALA |
| Release date | July 2025 |
| Modality | Text only |
Pricing
Open weights under Apache 2.0 — self-host at no cost.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| HumanEval | 95.1% | Benchgen evaluation | 2025-07 |
| MMLU-Pro | 67% | Benchgen evaluation | 2025-07 |
MiniCPM SALA vs Alternatives
| Model | HumanEval | MMLU-Pro | License |
|---|---|---|---|
| MiniCPM SALA | 95.1% | 67% | Apache 2.0 |
| Qwen2.5 VL 32B | 91.5% | 68.8% | Apache 2.0 |
MiniCPM SALA leads HumanEval (95.1% vs 91.5%). For MMLU-Pro: Qwen2.5 VL 32B slightly higher (68.8% vs 67%).
Related Models & Benchmarks
- Compare: Qwen2.5 7B Instruct
- Benchmarks: HumanEval · MMLU-Pro
Frequently Asked Questions
What is MiniCPM SALA?
OpenBMB's July 2025 Apache 2.0 model scoring 95.1% HumanEval, 67% MMLU-Pro. Excellent compact coding.Specs from OpenBMB's MiniCPM SALA release (July 2025) and Benchgen evaluations. Last updated 2026-07-24.