MiniCPM5 1B

Overview

MiniCPM5 1B is a state-of-the-art 1.08 billion parameter dense language model developed by OpenBMB, explicitly designed for on-device deployment, local execution, and agentic workflows. Built upon a standard LlamaForCausalLM backbone, MiniCPM5 1B offers an unprecedented balance of compact parameter scale, 128K native context length, and flexible dual-mode reasoning.

Architecture & Technical Specifications

Because MiniCPM5 1B adheres strictly to standard Llama model primitives, it requires no custom CUDA kernels and works out-of-the-box across all major LLM serving engines.

Key Capabilities & Features

1. Dual-Mode Reasoning (Think / NoThink): A single model checkpoint supports both instant low-latency responses ("NoThink") and extended Chain-of-Thought reasoning ("Think"). Users can toggle modes during prompt template application via enable_thinking.
2. Native 128K Context Length: Efficiently processes long documents, technical manuals, and multi-turn agent histories up to 131,072 tokens.
3. Agentic Tool Calling & Structured Output: Pre-trained and instruction-tuned for dependable function execution, API integration, and schema-constrained JSON generation.
4. Broad Engine Compatibility: Native support in vLLM, SGLang, llama.cpp (GGUF), Ollama, LM Studio, and MLX for Apple Silicon.

Benchmark Performance

MiniCPM5 1B sets a new benchmark for sub-2B parameter models:

• Average Benchmark Suite Score: 42.57 (significantly surpassing previous generation 1B SLMs averaging ~35.6).
• Artificial Analysis Intelligence Index: 17.9, ranking at the top of its parameter class.
• Specialized Strengths: Demonstrates exceptional capability on competition math (GSM8K/MATH) and code generation (HumanEval).

Quickstart Usage

PyTorch / Hugging Face Transformers

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "openbmb/MiniCPM5-1B"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Example 1: No-Think Mode (Fast response)
messages = [{"role": "user", "content": "What is the capital of France?"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

# Example 2: Think Mode (Chain-of-Thought reasoning)
messages = [{"role": "user", "content": "How many r's are in strawberry?"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

Ollama Deployment

ollama run minicpm5:1b

Verified References & Paper Links

• Hugging Face Model Weights: openbmb/MiniCPM5-1B
• GitHub Repository: OpenBMB/MiniCPM
• Technical Report: arXiv:2506.07900
• Foundational Paper: arXiv:2404.06395