WISEBENCH · COMMUNITY

Benchmark your local model

WiseBench runs a fixed suite of practical tasks — JSON output, reasoning, writing, coding, instruction following and more — against any OpenAI-compatible endpoint. Every model is scored with the same deterministic checks, and good or bad, the results are published here for comparison.

Run the suite from your machine

Start LM Studio, Ollama, llama.cpp or vLLM on the computer that hosts your model, then run the generated command in a terminal. The runner needs Python 3.10+ only; it talks to your endpoint locally and uploads one finished report.

LM Studio http://localhost:1234/v1 · Ollama http://localhost:11434/v1 · llama.cpp http://localhost:8080/v1. If your server needs a key, export WISEBENCH_MODEL_KEY first.


    

      
      Download runner
      
    

    
Loading suite details…

  


  

    Building your own runner?
    The same contract is available over HTTP: GET /api/suite returns the active task list, POST /api/results accepts one completed report (JSON), and GET /api/results returns everything published. The downloadable script is a small, readable reference implementation.
  


  

    
Model results

    
Loading submissions…