Human preference
LMArena Elo
Elo rating from blind pairwise votes by the public on LMArena. Measures what people prefer, not what is correct.
LMArena Elo is a rating, not a percentage 24 of the 159 models in this index report a score for it. The highest published figure here is 1439, from Gemini 2.5 Pro.
Reported scores
Top 20 on LMArena Elo
Ordered by the figure each maker published. Models that have not reported this benchmark are not listed — an absent score is not a low score.
| # | Model | Published score |
|---|---|---|
| 1 | Gemini 2.5 ProGoogle DeepMind | |
| 2 | Kimi K2 InstructMoonshot AI | |
| 3 | Llama 4 MaverickMeta AI | |
| 4 | Grok 3xAI | |
| 5 | Gemini 2.5 FlashGoogle DeepMind | |
| 6 | DeepSeek-R1DeepSeek | |
| 7 | Gemini 2.0 FlashGoogle DeepMind | |
| 8 | Qwen3-235B-A22BAlibaba Qwen | |
| 9 | Gemma 3 27BGoogle DeepMind | |
| 10 | DeepSeek-V3DeepSeek | |
| 11 | Command ACohere | |
| 12 | GPT-4oOpenAI | |
| 13 | Claude 3.5 SonnetAnthropic | |
| 14 | Llama 3.1 405BMeta AI | |
| 15 | Llama 3.3 70BMeta AI | |
| 16 | Qwen2.5-72BAlibaba Qwen | |
| 17 | GPT-4 TurboOpenAI | |
| 18 | Mistral Large 2Mistral AI | |
| 19 | Claude 3 OpusAnthropic | |
| 20 | Gemma 2 27BGoogle DeepMind |
Where these numbers come from
Every score on this page is a published figure, taken from the model's own card, system card, technical report or release post, or from a public leaderboard. CorX Labs did not run these evaluations. Most are self-reported by the lab that built the model, which means they were produced under that lab's own choice of prompt, scaffold and number of attempts — so treat them as a starting point for a shortlist, not as a settled ranking.
A score someone other than the model's maker measured is marked Independent and names its measurer. Those are the stronger numbers on this page — an outside harness has no reason to flatter anyone — and there are not many of them.
Where a figure has not been published, the cell reads Not reported rather than an estimate. Nothing here is inferred, interpolated or guessed. Each model records the month its row was last checked. Full method and caveats.
How to read this score
LMArena Elo comes from blind pairwise votes by the public, so it captures something no static benchmark can: whether people prefer the answer. It cannot be gamed by training on a test set, because there is no fixed test set.
Where it is weak
It measures preference, not correctness. Length, formatting and confident tone all reliably win votes, and voters are self-selected rather than representative. A model can climb by being agreeable rather than by being right.