# OpenGenerativeAI/llm-colosseum

Benchmark LLMs by fighting in Street Fighter 3! The new way to evaluate the quality of an LLM

Repository: https://github.com/OpenGenerativeAI/llm-colosseum
Canonical: https://ross.abutalabs.com/products/llm-colosseum
Homepage: https://llm-colosseum.phospho.ai
Language: Jupyter Notebook
License: MIT
License Family: permissive
Topics: genai, llm, benchmark, streetfighterai
Last push: 2025-03-21T10:35:26+00:00

## Health v2 (maintenance only)
Score: 28/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 12, release rhythm 28, longevity 63
- inputs: {"age_days": 893, "days_push": 530, "days_rel": 651, "gap_med": 41, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1484, forks 182 (observed 2026-08-28T04:04:51.344476+00:00)

## What it is
LLM Colosseum is a benchmarking tool that pits large language models against each other in real-time Street Fighter III matches to evaluate their decision-making quality. It produces an ELO-based leaderboard from hundreds of automated fights between models like GPT-4o, Pixtral, Claude, and Llama.

## Use cases
- benchmark llms by playing street fighter
- compare llm decision making in real time
- elo ranking of large language models
- evaluate llm agents in a game environment
- test which ai model is the best fighter
- alternative llm benchmark beyond static tests

## When to choose
- you want a novel, real-time evaluation of LLM reasoning and adaptability
- you want to compare vision and text modes of frontier models in an interactive setting
- you want an entertaining demo of LLM agent capabilities

## When to avoid
- you need rigorous, reproducible academic benchmarks
- you need standard NLP task evaluation like MMLU or code generation
- you need a production LLM evaluation pipeline

## Facets
- artifact type: application
- maturity: active
- function: benchmarking, llm-inference, agent-framework, machine-learning
- domain: large-language-models, artificial-intelligence, gaming-tools, analytics
- platform: python, browser
- tags: llm-evaluation, elo-ranking, street-fighter, game-benchmark, reinforcement-learning-comparison, jupyter-notebook, web, desktop

## Member repositories
- OpenGenerativeAI/llm-colosseum (main) score 28

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:51.344476+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:33:58.324344+00:00, confidence not recorded.
  - readme: https://github.com/OpenGenerativeAI/llm-colosseum (fetched 2026-08-28T04:04:51.344476+00:00, sha f830e0bc546d)
  - homepage: https://llm-colosseum.phospho.ai (fetched 2026-08-29T11:40:17.184252+00:00, sha a6873599d06f)
- Data as of 2026-08-30T08:39:29.467469+00:00.