StentorLabs/SLM-Arena-Matches
SLM Arena Matches Public records from SLM Arena. Each completed round has one JSON file under rounds/, named by a random round ID. The same file is updated when AI commentary or a human vote arrives. No sample rounds were inserted for setup. Records contain the prompt, response order, model names and repository IDs, generated outputs, the GPT OSS 120B commentary and parsed winner when available, and an optional human winner and comment. Winners are response labels (A through E);… See the full description on the dataset page: https://huggingface.co/datasets/StentorLabs/SLM-Arena-Matches.
SLM Arena Matches
Public records from SLM Arena. Each completed round has one JSON file under rounds/, named by a random round ID. The same file is updated when AI commentary or a human vote arrives. No sample rounds were inserted for setup.
Records contain the prompt, response order, model names and repository IDs, generated outputs, the GPT OSS 120B commentary and parsed winner when available, and an optional human winner and comment. Winners are response labels (A through E); ties and no-clear-winner decisions are stored as TIE and NO_CLEAR_WINNER. Missing decisions are null.
The Arena records completed rounds by default. Users can check Do not publish this round before running a match to opt out. The Arena does not intentionally collect account identifiers or IP addresses, but prompts and comments may contain information entered by users. Records are public and can be read or downloaded by anyone. Analytics count only valid model winner labels; ties and unavailable judgments do not create a model win.
The leaderboard reflects recorded rounds, not a controlled evaluation. Model selection, prompts, generation settings, missing outputs, and whether a human votes can all affect results. These counts should not be interpreted as a general model-quality ranking.