Arena Introduces Factuality Ranking
arena · x · 2026-07-16
Arena launched a new model ranking that weights human preference alongside factuality.
- Factuality is now live in Text Arena and Search Arena as an opt-in feature.
- It is calculated by sampling random battles, extracting verifiable claims from available web pages, and checking their accuracy.
- To support this, the team has annotated 2M+ claims from real LLM conversations, including 1.3M+ from Text Arena and 700K+ from Search Arena.
- Enabling factuality caused notable rank shifts in Text Arena: Claude Fable 5 dropped slightly to 2nd, GPT-5.5 climbed 13 spots to 7th, and Muse Spark fell from 7th to 20th.
- By provider, Meta saw the biggest drop, while Anthropic retained the overall top spot. Among open-source providers, Xiaomi improved the most, rising from 9th to 6th.
Related event: Arena Adds Factuality to Model Rankings(12 posts)→
More from Models
- Kimi K3 is "open source until you get successful": $20M revenue triggers extra license terms — alex_verem · 2026-10-27
- Kimi K3 Is 'Open' Until You Succeed: License Caps Free Use at $20M Revenue — alex_verem · 2026-10-27
- Screenshot shows 'GPT-6 Astra Ultra' in ChatGPT model picker — unverified leak or joke — justalexoki · 2026-10-27
- DeepSeek Safety Eval Finds "Over-Attribution": Model Mistakes Ordinary Tasks for Tests — Much_Boysenberry9124 · 2026-10-27
- Cohere flags popular open-source model attempting to upload probe file to Beijing bucket — stanislavfort · 2026-10-27
- MissionReport benchmark: frontier models score just ~60% on government status reporting — Exp_Mark · 2026-10-27