Leaderboard / OpenBMB (MiniCPM)

MiniCPM5 2B

Name: MiniCPM5 2B
Author: OpenBMB (MiniCPM)

OpenBMB (MiniCPM) · MiniCPM5 · released 2026-09-06 · data checked 2026-11-01

MiniCPM5 2B is a 2.5B-parameter dense model from OpenBMB (MiniCPM) (China), released 2026-09-06 under the Apache-2.0 license. It scores 57.5 on the Small LM Index, #7 of 21 tiny models. It needs about 2.4 GB of memory at 4-bit with an 8K-token context.

Small LM Index

57.5

#7 of 21 scored tiny models

Likely 55.8–59.2 · from 6 of 6 core benchmarks

Parameters

2.5B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

131K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 15 reported by OpenBMB (MiniCPM). How we label sources.

MMLU-Pro Lab-reported Knowledge70.8

Class best: Qwen3.5 4B 79.1

GPQA Diamond Lab-reported Knowledge70.2

Class best: Nanbeige4.2 3B 87.4

Humanity's Last Exam Lab-reported Reasoning8.9

Class best: Spark-X2.6 4B 17.9

AIME Lab-reported Math · AIME 202686.5

Class best: Spark-X2.6 4B 91.2

LiveCodeBench Lab-reported Coding · LiveCodeBench v669.1

Class best: LFM3 1.7B 73.5

SWE-bench Verified Lab-reported Coding46.4

Class best: Nanbeige4.2 3B 63.6

Terminal-Bench Lab-reported Agentic · Terminal-Bench 2.18.6

Class best: Nanbeige4.2 3B 44.1

τ²-Bench Lab-reported Agentic · Telecom domain only97.1

IFBench Lab-reported Instruction following66.3

Class best: Spark-X2.6 4B 75.8

MMLU Lab-reported Knowledge · MMLU-Redux84.7

IFEval Lab-reported Instruction following86.7

Class best: LFM3 1.7B 96.7

BFCL Lab-reported Agentic · BFCL v466.6

Class best: Spark-X2.6 4B 71.8

MATH-500 Lab-reported Math94.6

HMMT Lab-reported Math · HMMT Feb 202663.8

Class best: Nanbeige4.2 3B 82.8

AA-LCR Lab-reported Long context59.0

Class best: Spark-X2.6 4B 67.3

Specifications

Lab: OpenBMB (MiniCPM) (CN · China)
Architecture: Dense transformer
Total parameters: 2.5B
Active parameters: All (dense)
Size class: Tiny
Context window: 131K tokens
Input modalities: text
Reasoning: Hybrid (thinking on/off)
License: Apache-2.0
Openness: Open weights
Memory to run: ~2.4 GB at 4-bit, 8K context (more sizes)
Hugging Face: 1.1M downloads (30 days) · 1.8K likes

Weights on Hugging Face ↗

Compare with

MiniCPM5 2B vs LFM3 1.7B (1.7B, index 60.0)
MiniCPM5 2B vs LFM2.5 2.6B (2.7B, index 43.2)
MiniCPM5 2B vs Qwen3.5 2B (2.3B, index 33.0)
MiniCPM5 2B vs Granite 4.2 3B (3B, index 54.8)
MiniCPM5 2B vs Ministral 3 3B (3B, index 49.4)
MiniCPM5 2B vs SmolLM3 3B (3.1B, index 30.4)

Ranked in

Sources

https://huggingface.co/openbmb/MiniCPM5-2B

Something wrong here? Report it.

Run it locally

Uses bartowski/MiniCPM5-2B-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ2_M, IQ3_M, IQ3_XS, IQ3_XXS, IQ4_NL, IQ4_XS, Q2_K….

Ollama

ollama run hf.co/bartowski/MiniCPM5-2B-GGUF:Q4_K_M

llama.cpp

llama-server -hf bartowski/MiniCPM5-2B-GGUF:Q4_K_M

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

vLLM (GPU server)

vllm serve openbmb/MiniCPM5-2B

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	131K context
16-bit F16	5.8 GB	7.0 GB	11 GB
8-bit Q8_0	3.4 GB	4.6 GB	8.9 GB
4-bit Q4_K_M	2.3 GB	3.5 GB	7.7 GB

Check it against your GPU, Mac or phone →