MODELSBreaking
Arena’s interaction judge tests what agents actually finish
Arena open-sources ArenaWebdevPreferences and self-repair evals showing 20-point gains as Claude Fable 5.2 holds #1 at 30.48% win share.
Sara Moradi · 2 min read
§ Vigas Newsroom · Models
AI Models
Every model story in this section links directly to live Vigas leaderboard win share and real per-million-token pricing.
