ChatBot Comparison (Arena, formerly LMArena) compared with Artificial Analysis and Hugging Face Open LLM Leaderboard| ChatBot Comparison (Arena) | Artificial Analysis | Hugging Face Open LLM Leaderboard |
|---|
| Evaluation method: Rankings come from community votes on anonymous model pairs. A Bradley-Terry model, similar to Elo, turns the votes into ratings, with a style control factor. Only votes cast while model names are hidden count. | Evaluation method: It runs its own benchmarks and measurements. Its methodology covers intelligence, quality, performance and price for models, API endpoints and systems. | Evaluation method: Models were scored on six automated tests: IFEval, BBH, MATH, GPQA, MUSR and MMLU-PRO. No human preference is involved. |
| What is ranked: Leaderboards cover text, WebDev, vision, document, search, image and video generation, among others. The ranking reflects human preference, so it can favour answer style. | What is ranked: Includes language models, speech, image, video and music models. It also measures inference speed, time to first token, and price per token and per task. | What is ranked: Open-weight language models only, evaluated under a standardised protocol so results can be compared and reproduced. |
| Measurement limits: Preference votes are subjective and depend on who votes and which prompts they write. It does not report latency or price in the way an engineering benchmark does. | Measurement limits: Speed results use a common token unit, and reasoning token counts come from a set of 60 prompts. The page does not address independence or funding. | Measurement limits: The Space shown as running in the cached view, while a secondary source states that the leaderboard has since been archived. Check its current status before relying on it. |
| Cost and openness: The leaderboards can be viewed freely. Voting needs no purchase. | Cost and openness: Reading the published indices is free of charge on the site. Methodology pages describe how each index is built. | Cost and openness: Hosted by Hugging Face as a community Space, with results in public datasets. |
| Cuándo elegirlo: Choose it to see which chatbot people prefer for writing, coding or general chat, and to follow new releases. | Cuándo elegirlo: Choose it when speed, price and measured intelligence must be compared across providers for an engineering decision. | Cuándo elegirlo: Choose it for historical comparison of open-weight models on fixed benchmarks, mindful of its possible archived status. |
