An Anthropic model has been #1 on every day since July: six runs in 115 days, and Gemini 4 Carbon came closest, 1.9 points behind

An Anthropic model has led the ArtificialWatch bench index on every day since July 21, 2026, the first day on its record: 115 days, six runs at #1, four Claude models. No model from another lab has topped the index on any day. The closest was Google's Gemini 4 Carbon, 1.9 points behind Claude Opus 5.5 (96.2 to 98.1) on three recorded days from October 31 to November 3.

THE SIX RUNS

• Claude Fable 5: July 21 to 24, 4 days.
• Claude Opus 5: July 25 to August 4, 11 days. It took #1 the day after it went live on July 24.
• Claude Fable 5: August 5 to 13, 9 days.
• Claude Opus 5: August 14 to September 23, 41 days.
• Claude Opus 5.5: September 24 to November 8, 46 days, the longest single run. It went live on September 22.
• Claude Fable 5.2: since November 9, the day it went live, 4 days so far, at an index of 100.

In total Claude Opus 5 has held #1 for 52 days across its two runs, Claude Opus 5.5 for 46 and Claude Fable 5 for 13. Claude Sonnet 5.5 was #2 on 15 recorded days and never led.

THE CLOSEST CHALLENGERS

• Gemini 4 Carbon entered the top three on October 29, the day after Google announced it and 12 days before it reached Google's public API on November 10. It has been #2 on all 14 recorded days since. On November 11 it trailed Claude Fable 5.2 by 6.3 points (93.7 to 100).
• OpenAI reached #2 on two recorded days: GPT-5.6 on July 21 (91 to Claude Fable 5's 97.1) and GPT-6 Astra on October 28 (94.2 to Claude Opus 5.5's 100).
• On 17 recorded days, October 11 to 27, the whole top three was Anthropic.

WHAT IT MEANS

The index combines the public leaderboards, so leading it means ranking near the top of most boards at once, not winning each one: on November 5, six public leaderboards put four different models first. On that combined measure, each of Anthropic's last three flagship launches took #1 within two days of going live, and nothing from another lab has displaced one since the record began.

HOW WE COUNTED

The ArtificialWatch bench index is the median of a model's percentile ranks across the public leaderboards that rate it, rebuilt daily (artificialwatch.com/benchmarks). Its LMArena column takes each model's higher rating of LMArena's Text and WebDev boards, so its leader there can differ from the Text Arena's #1 (in LMArena's own data, read November 12, Claude Fable 5.2 led WebDev and Gemini 4 Carbon led Text). For each day we take the last index committed that day to the site's public history; 104 of the 114 days from July 21 to November 11 have one, and a day without one keeps the previous day's leader. Today's leader is read from the live index. A "recorded day" is one of those 104. Launch dates are the days ArtificialWatch verified each model live on a public API. The day-by-day record is at artificialwatch.com/best-ai-model/history, and the index is free to reuse under CC BY 4.0 (artificialwatch.com/data/bench-index.json).