Leaderboard / OpenBMB (MiniCPM)

MiniCPM-o 4.5

Name: MiniCPM-o 4.5
Author: OpenBMB (MiniCPM)

OpenBMB (MiniCPM) · MiniCPM-o · released 2026-02-03 · data checked 2026-11-02

MiniCPM-o 4.5 is a 9.4B-parameter dense model from OpenBMB (MiniCPM) (China), released 2026-02-03 under the Apache-2.0 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 7.6 GB of memory at 4-bit with an 8K-token context.

Omni model: sees images and video, hears audio and speaks back (full-duplex voice), from SigLIP2, Whisper-medium, CosyVoice2 and Qwen3-8B.

Small LM Index

—

Needs 2 core benchmarks (has 0)

Parameters

9.4B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

41K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best small model on that benchmark. Sources: 2 reported by OpenBMB (MiniCPM). How we label sources.

MMLU Lab-reported Knowledge · Instruct mode77.0

Class best: OLMo 3 7B Think 77.8

MATH-500 Lab-reported Math · Instruct mode77.0

Class best: Falcon H1R 7B 97.4

Specifications

Lab: OpenBMB (MiniCPM) (CN · China)
Based on: Qwen3 8B (fine-tune or distillation)
Architecture: Dense transformer
Total parameters: 9.4B
Active parameters: All (dense)
Size class: Small
Context window: 41K tokens
Input modalities: text, image, audio, video
Reasoning: Hybrid (thinking on/off)
License: Apache-2.0
Openness: Open weights
Memory to run: ~7.6 GB at 4-bit, 8K context (more sizes)
Hugging Face: 203.2K downloads (30 days) · 1.6K likes

Weights on Hugging Face ↗

Compare with

MiniCPM-o 4.5 vs MiniCPM-SALA (9.5B, index 52.6)
MiniCPM-o 4.5 vs Qwen3.5 9B (9B, index 63.9)
MiniCPM-o 4.5 vs Nemotron Nano 9B v2 (8.9B, index 52.5)

Sources

https://huggingface.co/openbmb/MiniCPM-o-4_5

Something wrong here? Report it.

Run it locally

Uses openbmb/MiniCPM-o-4_5-gguf (Q4_K_M). Swap the tag for another size: F16, Q4_0, Q4_1, Q4_K_M, Q4_K_S, Q5_0, Q5_1, Q5_K_M….

Ollama

ollama run hf.co/openbmb/MiniCPM-o-4_5-gguf:Q4_K_M

llama.cpp

llama-server -hf openbmb/MiniCPM-o-4_5-gguf:Q4_K_M

Then open http://localhost:8080 for a chat window.

vLLM (GPU server)

vllm serve openbmb/MiniCPM-o-4_5

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	41K context
16-bit F16	20 GB	24 GB	25 GB
8-bit Q8_0	11 GB	15 GB	17 GB
4-bit Q4_K_M	7.0 GB	11 GB	12 GB

Check it against your GPU, Mac or phone →