// MiniCPM / VoxCPM · Text LLM
MiniCPM5-2B: VRAM & local deployment
2.52B parameters. About 2.5 GB at 4-bit and 6.4 GB at BF16 with one 8K-token request.
released 2026-09-06licence: Commercial OKmemory figures are estimates
01Specs
| Publisher | OpenBMB · openbmb |
|---|---|
| Family / series | MiniCPM / VoxCPM · MiniCPM5 |
| Type | Text LLM (text-generation) |
| Parameters | 2.52B Source: counted from official safetensors weights |
| Context length | 128K tokens (131,072)max_position_embeddings in config.json |
| Architecture | 42 layers · GQA |
| Released weights | BF16 · 5.0 GBdownload |
| Licence | Apache-2.0 · Commercial OKPermissive licence (Apache-2.0 / MIT / CC-BY / OpenMDW): commercial use allowed. |
| Released | 2026-09-06 (Hugging Face repo created) |
02How much VRAM it needs (estimate)
Total = weights + 10% + 0.5GB overhead + KV cache for one request (FP16 KV). More concurrent users multiply the KV column.
| Precision | Weights | Total @ 8K | Total @ 32K | Total @ 128K | Smallest GPU | Mac |
|---|---|---|---|---|---|---|
| BF16 / FP16 (original) | 5.0 GB | 6.4 GB | 7.4 GB | 11.7 GB | 8GB GPU | Mac 16GB |
| FP8 / INT8 / Q8_0 | 2.7 GB | 3.8 GB | 4.8 GB | 9.1 GB | 8GB GPU | Mac 16GB |
| 4-bit (Q4_K_M / AWQ / GPTQ-Int4) | 1.5 GB | 2.5 GB | 3.6 GB | 7.8 GB | 8GB GPU | Mac 16GB |
KV cache per request: 8K 0.3 GB · 32K 1.4 GB · 128K 5.6 GB
Bits per weight: BF16 16, FP8 8.5, 4-bit 4.85. “Fits” = total ≤ 90% of memory; Mac GPU can use ~2/3 of unified memory (≤36GB) or ~3/4 (larger). For MoE, all experts must be in memory, so size by total parameters. Method →
03How to run it
Only runtimes the publisher or the runtime’s official library names for this model.
- vLLM — mentioned in the official model card
- SGLang — mentioned in the official model card
- llama.cpp — mentioned in the official model card
- Ollama — mentioned in the official model card
- LM Studio — mentioned in the official model card
- MLX — mentioned in the official model card
- Transformers — mentioned in the official model card
04Official repositories & quantised builds
5 official repo(s) from openbmb.
- MiniCPM5-2B main
- MiniCPM5-2B-MLX MLX
- MiniCPM5-2B-GGUF GGUF
- MiniCPM5-2B-Base Base
- MiniCPM5-2B-GPTQ GPTQ
05Other MiniCPM / VoxCPM models
All 23 →| Model | Params | 4-bit total | Released |
|---|---|---|---|
| MiniCPM5-1B · same series | 1.08B | 1.4 GB | 2026-05-21 |
| MiniCPM-V-4.7-35B-A3B | 35.2B / 3B | 24.2 GB | 2026-10-07 |
| MiniCPM-V-4.7-1B | 1.55B | 1.6 GB | 2026-10-06 |
| MiniCPM-V-4.6-Thinking | 1.3B | 1.5 GB | 2026-05-08 |
| MiniCPM-V-4.6 | 1.3B | 1.5 GB | 2026-04-13 |
| VoxCPM2 | 2.29B | 2.0 GB | 2026-04-03 |
| MiniCPM-SALA | 9.48B | 7.1 GB | 2026-02-11 |
| MiniCPM-o-4_5 | 9.37B | 8.0 GB | 2026-02-03 |
| VoxCPM1.5 | 802M | 1.0 GB | 2025-12-05 |
06Sources
Data checked 2026-09-29. Parameters, context and licence are copied from the official repository; memory figures are our estimates.