v1.2.0-alpha
Pre-release
New: Benchmark
A model leaderboard inside the app, replacing the navigation entry that used to
open an external website.
- Overview, leaderboard and comparison tabs
- Overall Elo plus Coding, Math, Creative Writing, Instruction Following,
Multi-Turn, Longer Query and Non-English - Search, organisation and open-weight filters, sorting, incremental loading
- Rank, percentile, vote count, confidence interval, licence and evaluation
date; Hugging Face metadata where available - Side-by-side comparison of up to four models
Data comes from the Parquet files in lmarena-ai/leaderboard-dataset on the
Hugging Face Hub. Snapshots are cached locally and refreshed daily.
News
Articles can be saved per user, anthropic.com is a new source, and the feed
screen has been reworked around provider and category filters.
Marketplace
Layout and filter work in the frontend, continuing the rework from 1.1.2.
Documentation
The README is now a documentation tree in English and German — installation,
architecture, features, privacy, transparency, troubleshooting, development,
roadmap and contribution guide. The privacy page lists every external
connection the application can make, including the ones that are easy to miss.
Platforms
Linux x64, macOS on Apple Silicon, Windows x64. Intel Macs are not built.
Alpha.