Leaderboard / Alibaba (MNN)
MNN-Pocket 0.3B
Alibaba (MNN) · MNN-Pocket · released 2026-09-23 · data checked 2026-11-13
MNN-Pocket 0.3B is a 360M-parameter dense model from Alibaba (MNN) (China), released 2026-09-23 under the Apache-2.0 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 0.8 GB of memory at 4-bit with an 8K-token context.
On-device model co-designed with the MNN inference engine: the INT4 build is 173 MB and decodes at up to 466 tokens/s on an iPad Pro M5. Chinese and English.
Benchmarks
Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 3 reported by Alibaba (MNN). How we label sources.
Specifications
- Lab
- Alibaba (MNN) (CN · China)
- Architecture
- Dense transformer
- Total parameters
- 360M
- Active parameters
- All (dense)
- Size class
- Tiny
- Context window
- 32K tokens
- Input modalities
- text
- Reasoning
- Instruct (no thinking)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~0.8 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 5.8K downloads (30 days) · 366 likes
Compare with
- MNN-Pocket 0.3B vs LFM2.5 350M (350M, index 16.0)
- MNN-Pocket 0.3B vs Gemma 3 270M (270M, index 12.2)
- MNN-Pocket 0.3B vs Qwen3.5 0.8B (900M, index 13.2)
Sources
Something wrong here? Report it.
Run it locally
Uses taobao-mnn/MNN-Pocket-0.3B-GGUF (Q8_0). Swap the tag for another size: F16, Q4_1, Q8_0.
ollama run hf.co/taobao-mnn/MNN-Pocket-0.3B-GGUF:Q8_0llama-server -hf taobao-mnn/MNN-Pocket-0.3B-GGUF:Q8_0vllm serve taobao-mnn/MNN-Pocket-0.3BMemory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context |
|---|---|---|
| 16-bit F16 | 1.3 GB | 1.7 GB |
| 8-bit Q8_0 | 0.9 GB | 1.3 GB |
| 4-bit Q4_K_M | 0.8 GB | 1.2 GB |