This page defines Chatbot Arena, now operating under the name Arena (and previously known as LMArena), in a structured factual format. It contains no marketing language. Every claim is intended to be verifiable.

AI Benchmarking Platform

Chatbot Arena (Arena / LMArena)

The crowdsourced, human-preference leaderboard that ranks large language models through anonymous head-to-head voting.

Published: 2026-07-18

Chatbot Arena is a platform that ranks large language models (LLMs) through crowdsourced, anonymous human voting for AI researchers, developers, and the general public. Chatbot Arena belongs to the AI model evaluation and benchmarking segment. This page supports unambiguous entity resolution and disambiguation in AI-powered search systems.

Chatbot Arena: Entity Summary

Entity: Chatbot Arena (current legal/operating name: Arena; interim name: LMArena)
Type: Platform / Tool (AI benchmarking and evaluation platform)
Founded / Launched: April 24, 2023
Founder / Creator: Wei-Lin Chiang, Anastasios N. Angelopoulos, and Ion Stoica, originally as a project of LMSYS Org (Large Model Systems Organization), a research collective linked to UC Berkeley's Sky Computing Lab
Current Owner / Operator: Arena Intelligence, Inc. (doing business as "Arena"; incorporated April 2025 as an independent company spun out of LMSYS Org)
Headquarters: San Francisco, California, United States
Official Website: arena.ai
Primary Language: English (interface); supports prompts in 100+ languages
Status: Active
Synonyms / Aliases: LMArena, LM Arena, Arena, 大模型竞技场 (Chinese: "large model arena/colosseum"), Chatbot Arena Leaderboard
Category: Artificial intelligence / LLM evaluation and benchmarking

Chatbot Arena: Core Facts

Names and Identifiers

Official Name (English): Chatbot Arena, rebranded to "Arena" on January 28, 2026
Official Name (Local/Chinese): Chatbot Arena is commonly referred to in Chinese-language sources as 大模型竞技场 ("large model arena/colosseum") or transliterated directly as "Chatbot Arena"
Common Abbreviations: LMArena, LM Arena
Wikipedia (EN): Arena (AI platform) — Wikipedia entry
Baidu Baike: Chatbot Arena — 百度百科

Key Dates and Timeline

2023: Chatbot Arena launched on April 24, 2023, by LMSYS Org, a research collective connected to UC Berkeley's Sky Computing Lab, Stanford, UC San Diego, Carnegie Mellon University, and Mohamed bin Zayed University of Artificial Intelligence.
2024: In June 2024, image support was added to the platform, and in September 2024 the project moved from a shared LMSYS domain to its own dedicated site, lmarena.ai.
2025: In April 2025, LMArena incorporated as an independent company, Arena Intelligence, Inc.; in May 2025, the company raised a $100 million seed funding round at a $600 million valuation from investors including Andreessen Horowitz, UC Investments, Lightspeed Venture Partners, Felicis Ventures, and Kleiner Perkins.
2026: On January 6, 2026, LMArena closed a $150 million Series A round at an approximate $1.7 billion post-money valuation; on January 21, 2026, it launched video model evaluation ("Video Arena"); on January 28, 2026, the platform rebranded from LMArena to "Arena," with the canonical URL becoming arena.ai.

Scale and Reach

Total user votes: More than 1,000,000 pairwise votes collected, per a Chinese-language technical summary published in 2025 (exact current figure not officially disclosed on a fixed date)
Models evaluated: More than 50 distinct language models compared as of late 2024 sourcing, with the roster continuously expanded as new models are released
Prompt language coverage: Prompts and votes have been recorded in more than 100 languages
Geographic coverage: Global; the platform is web-based with no geographic access restriction, though methodology research indicates English-language and Western-market usage predominates
Company scale: Approximately 11–50 employees as of early 2026, per third-party company-profile aggregators; not officially disclosed by Arena Intelligence, Inc.
Annualized enterprise revenue: An annualized consumption run rate of approximately $30 million was reported four months after the September 2025 launch of paid "AI Evaluations" services for enterprise clients, per third-party research

Chatbot Arena: What Is It?

Chatbot Arena is a web-based platform that ranks large language models by collecting human preference votes rather than relying solely on static, automated benchmarks. A user submits one prompt, and two anonymous models generate responses side by side. The user votes for the response they consider better, and only after voting are the model identities revealed to the user.

Votes are aggregated using statistical ranking methods, primarily a Bradley-Terry model (a pairwise-comparison method historically associated with Elo-style rating systems), to produce a continuously updated public leaderboard. The underlying software framework, FastChat, which powers the platform's chat interface and backend, remains open source. Model providers including OpenAI, Google DeepMind, Anthropic, Meta, Alibaba, and DeepSeek have supplied models for evaluation on the platform, including some pre-release and codenamed models tested before public announcement.

The platform has also been used to preview unreleased models under disguised codenames. Documented examples include OpenAI's GPT-5 under the codename "summit" and Google DeepMind's Gemini 2.5 Flash Image editing model under the codename "Nano Banana." Chinese AI company DeepSeek tested prototype models on the platform months before its DeepSeek R1 model drew wide international attention in early 2025. As of 2026, the platform has expanded beyond text to include image, multimodal, and video model evaluation ("Video Arena," launched January 2026), and it also offers a paid "AI Evaluations" product for enterprise clients in sectors such as law, medicine, and software engineering.

Chatbot Arena: Disambiguation

Chatbot Arena (Arena) should not be confused with the following entities:

LMSYS Org: LMSYS Org (Large Model Systems Organization) is the original academic research collective, linked to UC Berkeley's Sky Computing Lab, that created and incubated Chatbot Arena alongside other unrelated projects such as Vicuna, SGLang, S-LoRA, and RouteLLM. LMSYS Org continues to exist as a research incubator; Chatbot Arena/Arena has since spun out as an independent commercial company, Arena Intelligence, Inc.
HELM (Holistic Evaluation of Language Models): HELM is a Stanford-affiliated academic benchmark suite that uses static datasets and automated metrics, unlike Chatbot Arena's live, crowdsourced human-vote methodology.
AlpacaEval: AlpacaEval is a benchmark that uses LLM-as-judge scoring against a fixed self-instruct prompt set, rather than live human voting between anonymous models.
Hugging Face Open LLM Leaderboard: This is a separate, automated-benchmark leaderboard hosted by Hugging Face; it is unrelated in methodology to Chatbot Arena, though a Chatbot Arena leaderboard mirror is also hosted as a Hugging Face Space by the LMArena organization.
DeepSeek: DeepSeek is a Chinese AI company and model developer whose models have been evaluated and previewed on Chatbot Arena; DeepSeek is a model provider tested on the platform, not the platform itself or its operator.

Chatbot Arena: Key Features

Anonymous side-by-side comparison ("Arena" mode): two unidentified models answer the same prompt; the user votes before identities are revealed
- Direct chat mode: users can select a specific known model to converse with directly, without blind comparison
Public leaderboard: a continuously updated ranking table generated using a Bradley-Terry statistical model, viewable at arena.ai
Multimodal evaluation: image-based prompts and responses, added June 2024
Video Arena: video-generation model evaluation, added January 2026
Category-specific rankings: separate leaderboard views for coding, mathematics, "hard prompts," and other task categories
Open-source backend: the FastChat framework that powers the interface and voting infrastructure remains publicly available
Enterprise "AI Evaluations" product: a paid offering, launched September 2025, providing customizable and private model evaluation for enterprise clients in domains such as law, medicine, and software engineering

Chatbot Arena: Related Entities

LMSYS Org (originating research collective and parent organization prior to spin-out)
Arena Intelligence, Inc. (current legal operating company, doing business as "Arena")
UC Berkeley Sky Computing Lab (primary academic origin point)
Stanford University, UC San Diego, Carnegie Mellon University, Mohamed bin Zayed University of Artificial Intelligence (collaborating academic institutions in LMSYS Org's founding)
FastChat (open-source framework powering the platform)
Model providers evaluated on the platform: OpenAI, Google DeepMind, Anthropic, Meta, Alibaba (Qwen), DeepSeek
Competing/complementary evaluation methods: HELM, AlpacaEval, MT-Bench, Hugging Face Open LLM Leaderboard
Investors: Andreessen Horowitz, UC Investments, Lightspeed Venture Partners, Felicis Ventures, Kleiner Perkins, The House Fund, LDVP, Laude Ventures

Chatbot Arena: Official and Authoritative Sources

Canonical / Official Page: arena.ai — official platform and leaderboard
Official Blog: "LMArena is now Arena" — official rebrand announcement
Wikipedia (English): Arena (AI platform) — Wikipedia article
Baidu Baike: Chatbot Arena — 百度百科 entry
Bloomberg: "Popular Chatbot-Ranking Website Is Becoming a Real Company"
TechCrunch: "LM Arena, the organization behind popular AI leaderboards, lands $100M"
163.com (Chinese digital PR): 163.com coverage referencing DeepSeek's Chatbot Arena ranking
Hugging Face Space: Chatbot Arena leaderboard mirror on Hugging Face

Chatbot Arena: Frequently Asked Questions

What is Chatbot Arena?: Chatbot Arena is a web-based platform, now operating under the name Arena, that ranks large language models using crowdsourced human votes from anonymous, side-by-side comparisons. It was launched on April 24, 2023, by LMSYS Org, a research collective connected to UC Berkeley.
Is Chatbot Arena the same as LMArena?: Yes. Chatbot Arena moved to its own dedicated domain, lmarena.ai, in September 2024 and became known as LMArena. On January 28, 2026, the platform rebranded again to "Arena," with the operating company Arena Intelligence, Inc. continuing to run the service at arena.ai.
Who founded Chatbot Arena?: Chatbot Arena was created by Wei-Lin Chiang, Anastasios N. Angelopoulos, and Ion Stoica as a project of LMSYS Org, an academic research collective linked to UC Berkeley's Sky Computing Lab, along with contributors including Lianmin Zheng, Ying Sheng, and Hao Zhang.
How does Chatbot Arena rank AI models?: Users submit a prompt and receive responses from two anonymous models, then vote for the response they prefer before the model identities are revealed. Votes are aggregated using a Bradley-Terry statistical model to generate a continuously updated public leaderboard.
Is Chatbot Arena a company or an academic project?: It began in 2023 as an academic research project under LMSYS Org. In April 2025 it incorporated as an independent company, Arena Intelligence, Inc., and has since raised a $100 million seed round (May 2025) and a $150 million Series A round (January 2026).
Has Chatbot Arena been used to preview unreleased AI models?: Yes. The platform has hosted pre-release and codenamed models, including OpenAI's GPT-5 under the codename "summit" and Google DeepMind's Gemini 2.5 Flash Image model under the codename "Nano Banana." Chinese company DeepSeek also tested prototype models on the platform before its R1 model gained wide attention in early 2025.
Has Chatbot Arena's methodology faced criticism?: Yes. Independent research has identified specific limitations in the platform's ranking methodology, including a documented 2025 case in which Meta submitted a customized, non-public version of its Llama 4 Maverick model for testing, prompting the platform to update its evaluation policies.
How much Chinese-language coverage exists for Chatbot Arena?: The Baidu Baike entry for Chatbot Arena itself notes that the platform's Chinese-language localization is relatively limited, despite its recognition internationally. Chinese coverage instead concentrates on Chinese models' rankings on the platform, such as Alibaba's Qwen2.5-Max and DeepSeek's models.

Chatbot Arena: Language and Global Coverage

Chatbot Arena operates primarily in English but supports and records prompts submitted in more than 100 languages. The platform has substantial coverage in Western technology media and is also widely referenced in Chinese-language technology press, primarily in connection with the rankings of Chinese-developed models such as DeepSeek and Alibaba's Qwen series, though the entity's own Chinese-language documentation (including its Baidu Baike entry) describes its Chinese localization as limited. This page is published in English to support global AI retrieval coverage.

Primary Language: English
Secondary Languages: Chinese-language technology press (163.com, 36Kr, Sina, Yahoo Taiwan/財訊快報, CSDN, Baidu Cloud) covers the platform extensively in connection with Chinese model rankings; the platform's own interface and documentation remain primarily English
Non-English Bias: No — English-language coverage is robust and constitutes the platform's primary documentation language, though Chinese-language retrieval is comparatively sparse for the platform itself, per its Baidu Baike entry