Arena.ai

Formerly LMArena, Arena.ai (UC Berkeley origins) runs crowdsourced, blind head-to-head model evaluations across text, code, image, and video. Rankings use Bradley-Terry scoring over 250M+ real conversations. Offers commercial eval services for labs and enterprises, plus Max, a community-vote-powered model router.

Last 12 months

Focus areas

BENCHMARKS•IMAGE•LLMS•VIDEO GENERATION•VIDEO UNDERSTANDING•VISION LANGUAGE

Links

Agent Arena Splits Into 3 Categories, Proving No Single Model Wins

Arena.aiBenchmarks

Aug 17

LM Arena Proves Claude Opus 5 Writes 3x More but Simpler

Aug 12

Microsoft's MAI-Image-2.6 Jumps to #2, Beating GPT Image 2 in 3D

Aug 10

UCLA Finds AI Reward Hack Monitors Collapse to 28% on Real Cheating

Arena.aiPost Training

Aug 08

LM Arena's AutoEval Ranks AI Models in Hours, Beating Frontier LLM Judges

Arena.aiBenchmarks

Jul 30

LM Arena Breaks Down 52 Image Models Across 7 Categories With 28M Votes

Jul 30

LM Arena's Factuality Leaderboard Reveals OpenAI Leads While Meta Falls 13 Spots

Arena.aiBenchmarks

Jul 15

Arena.ai's Code Arena Now Builds and Ships Real Fullstack Apps

Arena.aiVibe Coding

Jul 02

ByteDance's Seed 2.1 Pro Matches Claude Opus 4.6 on Frontend Code Before Launch

Arena.aiDevelopment

Jun 22

LM Arena Ships Agent Mode to Rank AI Models on Real Multi-Step Tasks

Jun 04