Leaderboard / OpenBMB (MiniCPM)
MiniCPM-SALA
OpenBMB (MiniCPM) · MiniCPM · released 2026-02-11 · data checked 2026-11-02
MiniCPM-SALA is a 9.5B-parameter dense model from OpenBMB (MiniCPM) (China), released 2026-02-11 under the Apache-2.0 license. It scores 52.6 on the Small LM Index, #10 of 26 small models. It needs about 6.7 GB of memory at 4-bit with an 8K-token context.
Long-context model mixing sparse (25%) and linear (75%) attention for million-token contexts; up to 3.5x faster than Qwen3-8B at 256K tokens.
Benchmarks
Each bar shows this model's score. The tick marks the best small model on that benchmark. Sources: 4 reported by OpenBMB (MiniCPM). How we label sources.
Not reported: estimated
What the index model expects from this model's other scores. Not a measurement.
Specifications
- Lab
- OpenBMB (MiniCPM) (CN · China)
- Architecture
- Dense transformer
- Total parameters
- 9.5B
- Active parameters
- All (dense)
- Size class
- Small
- Context window
- 1M tokens
- Input modalities
- text
- Reasoning
- Hybrid (thinking on/off)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~6.7 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 3.7K downloads (30 days) · 686 likes
Compare with
- MiniCPM-SALA vs Qwen3.5 9B (9B, index 63.9)
- MiniCPM-SALA vs Nemotron Nano 9B v2 (8.9B, index 52.5)
- MiniCPM-SALA vs Apertus 1.5 8B (8.9B, index 20.1)
- MiniCPM-SALA vs DeepSeek-R1-0528-Qwen3-8B (8.2B, index 53.1)
Sources
Something wrong here? Report it.
Run it locally
No GGUF build from a well-known publisher yet.
vllm serve openbmb/MiniCPM-SALAMemory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 256K context |
|---|---|---|---|
| 16-bit F16 | 20 GB | 21 GB | 28 GB |
| 8-bit Q8_0 | 11 GB | 12 GB | 19 GB |
| 4-bit Q4_K_M | 6.6 GB | 7.5 GB | 15 GB |