Crowdsourced model evaluation platform (formerly LMSYS Chatbot Arena). Maintains the Text Arena, WebDev Arena, Arena-Hard and MT-Bench. On LLM Stats it maintains 5 benchmarks, led by Arena Hard, Arena-Hard v2 and MT-Bench. It ships no models of its own; each score on its benchmarks carries its own provenance label on the model page.

Which LMArena benchmarks count toward the LLM Stats Score?

Arena Hard, Arena-Hard v2, MT-Bench, LMArena Text Leaderboard and WebDev Arena count toward the LLM Stats Score. Only benchmarks admitted to the core set move the main leaderboard; the rest are shown for transparency.

Which models lead LMArena's benchmarks?

Qwen3 235B A22B leads Arena Hard at 95.6% (27 models scored), MiMo-V2-Flash leads Arena-Hard v2 at 86.2% (19 models scored) and Hermes 3 70B leads MT-Bench at 9.0% (13 models scored). Each benchmark page lists the full ranking and how the results were produced.

Where are LMArena's official benchmark results?

LMArena's official site is https://lmarena.ai. Each benchmark page on LLM Stats links to that benchmark's own paper or leaderboard, and shows the scores next to every other benchmark so models can be compared on one page.

LMArena AI Benchmarks: Arena Hard, Arena-Hard v2

Maintained by LMArena

5 benchmarks

Arena Hard
Arena-Hard-Auto is an automatic evaluation benchmark for instruction-tuned LLMs consisting of 500 challenging real-world prompts curated by BenchBuilder. It includes open-ended software engineering problems, mathematical questions, and creative writing tasks. The benchmark uses LLM-as-a-Judge methodology with GPT-4.1 and Gemini-2.5 as automatic judges to approximate human preference. Arena-Hard achieves 98.6% correlation with human preference rankings and provides 3x higher separation of model performances compared to MT-Bench, making it highly effective for distinguishing between models of similar quality.
chat reasoning general creativity writing Counts toward the LLM Stats Score
Leader
Qwen3 235B A22B95.6%
Models scored
27
Arena-Hard v2
Arena-Hard-Auto v2 is a challenging benchmark consisting of 500 carefully curated prompts sourced from Chatbot Arena and WildChat-1M, designed to evaluate large language models on real-world user queries. The benchmark covers diverse domains including open-ended software engineering problems, mathematics, creative writing, and technical problem-solving. It uses LLM-as-a-Judge for automatic evaluation, achieving 98.6% correlation with human preference rankings while providing 3x higher separation of model performances compared to MT-Bench. The benchmark emphasizes prompt specificity, complexity, and domain knowledge to better distinguish between model capabilities.
reasoning general creativity writing Counts toward the LLM Stats Score
Leader
MiMo-V2-Flash86.2%
Models scored
19
LMArena Text Leaderboard
LMArena Text Leaderboard is a blind human preference evaluation benchmark that ranks models based on pairwise comparisons in real-world conversations. The leaderboard uses Elo ratings computed from user preferences in head-to-head model battles, providing a comprehensive measure of overall model capability and style.
chat reasoning general Counts toward the LLM Stats Score
Leader
Grok-4.1 Thinking1483
Models scored
2
MT-Bench
MT-Bench is a challenging multi-turn benchmark that measures the ability of large language models to engage in coherent, informative, and engaging conversations. It uses strong LLMs as judges for scalable and explainable evaluation of multi-turn dialogue capabilities.
chat reasoning roleplay general communication creativity Counts toward the LLM Stats Score
Leader
Hermes 3 70B9.0%
Models scored
13
WebDev Arena
WebDev Arena is a leaderboard for evaluating AI models on web development tasks, including zero-shot generation, complex prompts, and interactive web UI creation. Models are ranked using Elo ratings based on their performance in coding and web development challenges.
reasoning frontend development code Counts toward the LLM Stats Score
Leader
Gemini 4 Carbon1768
Models scored
2

In one line

LMArena maintains Arena Hard (counts toward the LLM Stats Score, 27 models scored), Arena-Hard v2 (counts toward the LLM Stats Score, 19 models scored), LMArena Text Leaderboard (counts toward the LLM Stats Score, 2 models scored), MT-Bench (counts toward the LLM Stats Score, 13 models scored) and WebDev Arena (counts toward the LLM Stats Score, 2 models scored). It ships no models of its own on LLM Stats.

LMArena: AI benchmarks & research