Model library / openbmb
MiniCPM5-2B is a 2-billion-parameter dense transformer aimed at on-device and resource-constrained deployments.
Parameters in total
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
MiniCPM5-2B is a dense 2-billion-parameter transformer released by OpenBMB. The publisher describes it as a compact model for on-device, local deployment and resource-constrained environments, extending the MiniCPM5 series with improved coding, mathematics, long-context understanding, tool use and agentic tasks, and demonstrates competitive performance against larger open-source models.
The model contains 2.5 billion parameters and uses bfloat16 weights. It supports a context window of 131 072 tokens, enabling very long inputs. Distributed under the Apache-2.0 licence, it can be integrated into commercial or research pipelines without additional restrictions. Its dense architecture without expert layers simplifies runtime optimisation on CPUs and GPUs, and the large context length is useful for document-level tasks.
Specification
| Repository | openbmb/MiniCPM5-2B |
| Publisher | openbmb |
| Published | 6 September 2026 |
| Architecture | LlamaForCausalLM |
| Model type | llama |
| Parameters | 2,516,756,480 |
| Layers | 42 |
| Hidden size | 2,048 |
| Attention heads | 16, 2 key/value heads |
| Context length | 131,072 tokens |
| Vocabulary | 130,560 tokens |
| Weight format | bfloat16 |
| Licence | apache-2.0 |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 4.7 GiB | Full precision as released |
| fp8 | 2.3 GiB | 8-bit, near-lossless on most models |
| int4 | 1.2 GiB | 4-bit, smallest footprint |
The cache costs 42 KiB per token at 16-bit, so the full 131,072-token context of one request needs about 5.3 GiB. Concurrency multiplies that number, not the weights.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 1 | 14.4 GiB | the full 131,072 tokens |
| RTX 5090 32GB | 1 | 21.6 GiB | the full 131,072 tokens |
| L40S 48GB | 1 | 36.0 GiB | the full 131,072 tokens |
| A100 80GB | 1 | 64.8 GiB | the full 131,072 tokens |
| H100 80GB | 1 | 64.8 GiB | the full 131,072 tokens |
| RTX PRO 6000 96GB | 1 | 79.2 GiB | the full 131,072 tokens |
| H200 141GB | 1 | 119.7 GiB | the full 131,072 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve openbmb/MiniCPM5-2B \ --max-model-len 131072 \ --gpu-memory-utilization 0.90
Questions
The weights take about 5 GiB at the released precision. That fits on one RTX 4090 24GB, which leaves roughly 14 GiB for the key-value cache.
The repository declares apache-2.0. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 131,072 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email [email protected] with the model name and roughly what volume you expect.
Also in the library
MiniCPM5-2B is a 2.5-billion-parameter dense Llama-style causal language model released by openbmb.
MiniCPM5-2B-SFT is a 2.5 billion-parameter dense Llama-style transformer released by OpenBMB for on-device, resource-constrained …
MiniCPM5-2B is a 2.5 billion-parameter dense Llama-style causal model from openbmb for on-device, resource-constrained deployment.
NeoHorse-1-4B is a 4.2B open-weight model published by TokenRhythm on Hugging Face. It uses the Qwen3_5ForCausalLM architecture w…
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.