Keelemudelite mõõdupuu

Independent LLM leaderboard for Estonian, published by Eesti Keele Instituut (Institute of the Estonian Language). The interactive site requires JavaScript; the source data is published openly and listed below.

Sõltumatu hindamisraamistik suurte keelemudelite võimekuse mõõtmiseks eesti keeles. Allikandmed on avalikud ja loetletud allpool.

Source data / Allikandmed

• /llms.txt — agent on-ramp: every data file with raw URLs and the aggregation rule
• summary.csv — one row per model: id, provider, release date, overall, per-benchmark scores
• models.json — model metadata (id, name, provider, ISO release date, tags)
• benchmarks.json — benchmark registry with bilingual names, tags, and links to per-benchmark Markdown
• results.json — nested per-model scores including overall and category breakdowns
• results.jsonl — one JSON record per (model, benchmark) run
• leaderboard-data-ui — full data repository with README, schema, and per-benchmark Markdown descriptions

Quick path for agents: fetch summary.csv — it joins model metadata, release dates, and all benchmark scores into one table. The aggregation rule and schema are documented in the data repo README.