@mylifcc: http://Arena.ai just officially added 'Factuality' to model rankings. The leaderboard now supports weighted viewing of 'Human Preference + Factuality' (default 25% factuality weight). They have annotated over 2 million claims from real conversations (Text Ar…

X AI KOLs Timeline Tools

Summary

Arena.ai has added Factuality to model rankings, supporting weighting of human preference and factuality, and showing changes in model rankings.

http://Arena.ai has officially added 'Factuality' to model rankings. The leaderboard now supports weighted viewing of 'Human Preference + Factuality' (default 25% factuality weight). They have annotated over 2 million claims from real conversations (1.3 million+ from Text Arena, 700k+ from Search Arena), extracting verifiable claims through random sampling battles and verifying accuracy. Factuality is now available in Text and Search Arena (can be manually enabled). Key changes after enabling Factuality: GPT-5.5 rose the most, +13 spots to 7th place Claude Fable 5 dropped slightly, still at 2nd Muse Spark fell the most, from 7th to 20th Meta dropped from 2nd to 5th, Anthropic remains 1st Among open-source models, Xiaomi improved the most (9th → 6th) Interestingly, only OpenAI consistently improved factuality over a longer period; most open-source models saw their scores drop when factuality weight increased (with a few exceptions like mistral-medium-3.5 and Tencent huyuan-hy3-preview). In Search Arena, GPT-5.5-search topped the Factuality leaderboard, and GPT-5.2-search also rose significantly. Detailed methodology, interactive charts under different weights, and full analysis: https://arena.ai/blog/factuality-in-arena/… This update turns 'telling the truth' from an implicit requirement into a quantifiable ranking dimension, providing reference value for agent systems that need reliable outputs.
Original Article
View Cached Full Text

Cached at: 07/16/26, 06:06 AM

http://Arena.ai has officially added “Factuality” to its model rankings.
The leaderboard now supports weighted viewing of “Human Preference + Factuality” (default 25% factuality weight). They’ve labeled claims from over 2 million real conversations (1.3M+ in Text Arena, 700K+ in Search Arena), extracting verifiable claims through random sampling of battles and verifying accuracy.
Factuality is now live in Text Arena and Search Arena (can be manually enabled).

Key changes after enabling Factuality:

  • GPT-5.5 rose the most, up 13 spots to #7
  • Claude Fable 5 dropped slightly, still at #2
  • Muse Spark fell the most, from #7 to #20
  • Meta dropped from #2 to #5, Anthropic remains #1
  • Among open-source models, Xiaomi improved the most (#9 → #6)

Interestingly, only OpenAI has consistently improved factuality over a longer period; most open-source models actually score lower when factuality weight is increased (with few exceptions like mistral-medium-3.5 and Tencent’s huyuan-hy3-preview).

In Search Arena, GPT-5.5-search takes the Factuality top spot, and GPT-5.2-search also climbed significantly.
Detailed methodology, interactive charts for different weights, and full analysis:
https://arena.ai/blog/factuality-in-arena/…

This update makes “telling the truth” a quantifiable ranking dimension, shifting it from an implicit requirement — valuable for agent systems that need reliable outputs.


Arena AI: The Official AI Ranking & LLM Leaderboard

Source: https://arena.ai/
Inputs are processed by third-party AI and responses may be inaccurate.

Your conversations and certain other personal information will be disclosed to the relevant AI providers and may otherwise be disclosed publicly to help support our community and advance AI research.

Do not submit to our Services any personal information or other sensitive information that you would not want to be shared publicly. By continuing to use our Services, you acknowledge and direct us to engage in such sharing.

Arena.ai (@arena):
Introducing factuality in the Arena: a new ranking of models according to a weighted combination of human preference and factuality.

Model rankings are now viewable according to a weighted combination of human preference and factuality. Factuality is live in our Text and Search

Similar Articles

@MindfulReturn: Today I saw an interview with Professor Huang Biwei (@huang_biwei) and learned about their new round of funding! After learning about the Aether AI solution and taking a closer look at their direction, let me share my thoughts: The next paradigm of AI is not bigger models, but causality. 1. Correlation Ceiling: Why the visuals are...

X AI KOLs Timeline

This article offers an in-depth analysis of the Causal World Model (CWM) proposed by Aether AI (原识之智), arguing that the next AI paradigm will shift from correlation to causation. It discusses the theoretical foundations, technical architecture, and potential impact on video generation and embodied intelligence.

@FinanceYF5: How do regular users use AI in their personal lives? The latest analysis based on approximately 40,000 Claude conversations shows that health and wellness is the largest category, followed by career and professional development. The third category (relationships) shows a significant drop compared to the top two.

X AI KOLs Following

An analysis of approximately 40,000 Claude conversations reveals that the primary ways regular users employ AI in their personal lives are for health and wellness, followed by career development, with health-related queries making up the largest proportion.

RealityTest: How People Probe AI Identity and Whether Models Disclose It

arXiv cs.CL

This paper introduces RealityTest, a multimodal, multilingual benchmark to evaluate whether AI systems disclose their identity when probed by users, based on real human queries collected across 49 countries. It finds that only 31% of people ask directly about identity, and that human questions are more diverse than synthetic ones, revealing that phrasing and context matter more for disclosure than the specific model.

Natural Conversations with GPT-Live

YouTube AI Channels

OpenAI showcases breakthrough in personalization and naturalness of the new voice model, capable of natural brainstorming conversations, displaying empathy and timely interjection, approaching human conversation rhythm.