MiniCPM5-2B
MiniCPM5-2B by OpenBMB is a compact dense MiniCPM5-series LLM with RL post-training, one-prompt deploy/fine-tune Agent Skills and a Persona LoRA Hub.
Model Overview
Capabilities, design details, and architectural traits
MiniCPM5-2B - a compact dense model in the MiniCPM5 series
MiniCPM5-2B belongs to the MiniCPM5 series from OpenBMB, a line of compact dense LLMs built to push strong performance into small model sizes. The series is shaped by RL post-training.
What most clearly separates the MiniCPM5 series is its packaging around Agent Skills. Two top-level router skills, minicpm5-deploy and minicpm5-finetune, let a user start from a single prompt and get routed to the right backend commands, instead of hand-assembling inference or training setups.
| Trait | Detail |
|---|---|
| Series identity | Member of the MiniCPM5 series of compact dense LLMs from OpenBMB |
| Post-training | Documented RL post-training |
| Agent Skills | One-prompt minicpm5-deploy and minicpm5-finetune router skills, part of a 16-skill set |
| Deployment targets | Cookbooks for transformers, vLLM, SGLang, AWQ, GPTQ, llama.cpp, Ollama, LM Studio and MLX |
| Fine-tuning paths | Cookbooks for TRL, LLaMA-Factory, ms-swift, unsloth and xtuner |
| Community layer | A Persona LoRA Hub with a contributor guide, plus a Desktop Pet concept |
One-prompt deployment story
The MiniCPM5 project leads with the idea that a small model should be deployable or fine-tunable from one prompt. The router skills dispatch to the underlying cookbook commands for each backend, so the detailed setup lives below the fold rather than in the user's way.
Persona LoRA Hub
Beyond the base model, the series ships a community Persona LoRA Hub, inviting contributors to build and share persona adapters rather than baking personalities into the model itself.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | OpenBMB | Anthropic | Anthropic |
| Release Date | September 7, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 262K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $4 | $2 |
| Output Pricing | Free Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | textimage | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 12.5 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 14.5 | - | - |
| Agentic Index | 6.7 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Top AI Models
Leading alternatives by intelligence score
