Three AI coding leaderboards, three different leaders, and the model that tops ArtificialWatch's coding index wins none of them

Three public AI coding leaderboards, read on November 5, each put a different model first among the models ArtificialWatch tracks, and none of them is the model that leads ArtificialWatch's coding index. That is Claude Opus 5.5, which is second on LMArena's coding board and on LiveBench's, and 5th on Vals' Vibe Code Bench.

WHO LEADS WHERE

- LMArena Coding: Gemini 4 Carbon, which is not on a public API yet, ahead of Claude Opus 5.5.

- LiveBench Coding: Claude Fable 5.1, ahead of Claude Opus 5.5.

- Vals Vibe Code Bench: Claude Sonnet 5.5, ahead of Gemini 4 Argon and Claude Haiku 5.5.

WHY THE INDEX DISAGREES

The coding index converts each board's score to a percentile and takes the median, so it rewards a model that places high on all three over one that wins a single board. Claude Opus 5.5 leads it at 96, Gemini 4 Argon is second at 95, and Gemini 4 Carbon third at 92.4 on the two boards that rate it.

HOW WE COUNTED

From ArtificialWatch's coding leaderboard (artificialwatch.com/coding, also at /data/coding-index.json): LMArena's coding board, the mean of LiveBench's Coding and Agentic Coding categories, and Vals' Vibe Code Bench, read November 5, 2026. Ranks are among the models ArtificialWatch tracks; a board can also rate models we don't.