Claude Opus 4.6 takes No. 1 after LMArena adds factuality
LMArena added a model ranking based on a combining human preference with factuality. It measures factuality by taking random model comparisons, finding claims that can be checked on the web, verifying them, and comparing each model’s average correctness. More than 2 million claims from real conversations have been labeled, including over 1.3 million from Text Arena and over 700,000 from Search Arena. With factuality included, reached No.
1 in the text ranking, while slipped slightly to No. 2. GPT-5.5 made the biggest gain, rising 13 places to No. 7.
Muse Spark had the biggest fall, dropping 13 places from No. 7 to No. 20. GPT-5.5-search rose one place in the search ranking, and the factuality setting must be turned on manually in both arenas.
Key points
- The new combines human preference with factuality.
- The measurement uses more than 2 million checked claims from real conversations.
- reached No. 1 in the factuality-enabled text ranking.
- GPT-5.5 rose 13 places to No. 7, while Muse Spark fell 13 places to No. 20.
- Factuality is available in both text and search rankings but is not enabled by default.