Data Desk · Updated weekly

AI Leaderboard

Sort by what you actually use the model for· Scores via LLM Stats, as of —

		ⓘ	ⓘ	ⓘ	ⓘ	ⓘ	ⓘ	ⓘ	ⓘ	ⓘ
1	GPT-5.1 OpenAI	–	92.0	88.0	–	–	–	–	–	–
2	Claude Sonnet 4.5 Anthropic	–	91.0	90.0	–	–	–	–	–	–
3	Gemini 2.5 Pro Google	–	88.0	84.0	–	–	–	–	–	–
4	ERNIE-X 350B Baidu	–	84.0	80.0	–	–	–	–	–	–
5	Qwen3-Max Alibaba	–	83.0	85.0	–	–	–	–	–	–
6	Llama 4 405B Meta	–	80.0	78.0	–	–	–	–	–	–
7	Mistral Large 3 Mistral	–	78.0	77.0	–	–	–	–	–	–
8	DeepSeek-V4 DeepSeek	–	82.0	83.0	–	–	–	–	–	–

TrueSkill conservative rating · higher is better · Data: LLM Stats

Methodology

Each model's score is a TrueSkill rating - the same system Xbox uses to rank players. Every published benchmark result counts as a head-to-head match between models, and beating a strong model raises a rating more than beating a weak one. We show the conservative estimate: a floor the model is about 99% likely to clear, so models with only a few benchmark results score lower until more evidence arrives. Ratings are grouped by category (General, Reasoning, Coding, Math, Search, Vision), compiled by LLM Stats, and refreshed here every Monday.