lmarena.ai is the former official website of LMArena, an AI model comparison and ranking platform now operating as Arena at arena.ai, where users can compare AI-generated responses, vote for their preferred models, and explore performance leaderboards.
What Is lmarena.ai, and Why Did Its Website Change?
lmarena.ai was the official domain of LMArena, a platform developed from the Chatbot Arena research project associated with researchers at the University of California, Berkeley.
Its purpose is to evaluate artificial intelligence models using feedback from real people. Instead of relying exclusively on standardized benchmark tests, the platform collects preferences from users comparing AI responses to practical questions and tasks.
On January 28, 2026, LMArena officially announced its rebranding to Arena. The company explained that its evaluation activities had expanded beyond large language models into image generation, video, coding, and other AI capabilities.
As of October 31, 2026, visiting lmarena.ai redirects to arena.ai. This is an official domain transition, not evidence that the service has shut down or been replaced by an unrelated website.
Arena Intelligence Inc. operates the current platform.
How Does Arena Compare AI Models?
Arena primarily compares AI models through head-to-head evaluations in which users decide which response they prefer.
Its best-known evaluation method is Battle Mode, which works as follows:
- A user submits a question, instruction, or creative task.
- Arena sends the prompt to two participating AI models.
- The models generate responses without initially revealing their identities.
- The user compares the results and votes for the preferred answer, a tie, or neither response.
- Arena reveals the model identities after the vote.
The anonymous comparison helps reduce the influence of brand recognition on user preferences.
According to Arena's official methodology, qualifying votes contribute to public rankings through a statistical system called Bradley–Terry, which estimates relative model performance using paired comparisons.
Arena also offers Direct Mode for interacting with a selected model and Side by Side Mode for comparing specific models.
However, votes from these non-anonymous modes do not contribute to the official public leaderboard.
What Features Does Arena Offer?
Arena supports several AI evaluation categories, allowing users to explore models according to the tasks they perform.
Its available capabilities include:
- Text Arena: Comparing AI responses for writing, reasoning, explanations, and general questions.
- Code Arena: Evaluating programming and web development capabilities.
- Image Arena: Comparing image generation and editing results.
- Video Arena: Evaluating AI-generated video outputs.
- Vision Arena: Testing models that interpret visual information.
- Search Arena: Comparing AI systems that retrieve and synthesize information.
- Document Arena: Evaluating document understanding and analysis.
- Agent Arena: Examining AI systems that perform more complex, multi-step tasks.
Not every feature supports identical interaction modes or model selections.
Arena also develops Max, an AI routing system designed to select models according to task requirements and performance considerations.
In October 2026, Arena introduced an updated agentic Max router intended to balance model performance with execution costs.
These developments show how the platform has expanded from chatbot comparisons into broader AI evaluation.
How Can You Use the Arena Leaderboard to Find Better AI Models?
The Arena leaderboard helps users identify models that perform well in particular evaluation categories.
Visitors can open the leaderboard, select a category, and examine the available rankings.
Important measurements include:
- Rank: A model's position relative to other evaluated models.
- Score: A statistical measurement derived from evaluation results.
- Votes: The number of comparisons contributing to a model's evaluation.
- Rank Spread: The estimated range of possible positions, reflecting statistical uncertainty.
Some categories also display model pricing information or other performance measurements.
A model that ranks highly for general conversation may not perform equally well in coding, image generation, or specialized research.
For example, a developer choosing an AI coding assistant should examine coding-related evaluations instead of relying entirely on overall text rankings.
Arena also provides specialized filters, including evaluation categories associated with professional tasks.
The important distinction is that rankings measure performance under particular evaluation conditions. They are useful comparison signals, not guarantees that one model will always produce the most accurate answer.
Is lmarena.ai Free to Use in 2026?
Arena provides free access to AI comparison features, but usage is subject to limits.
Its official help center describes a credit-based usage system in which users receive a daily allocation of free usage credits.
The 24-hour usage period begins when a user submits their first prompt. Credits are consumed during applicable interactions, and the interface displays information as the available balance approaches its limit.
Once the balance is exhausted, users must wait for their credits to refresh.
Importantly, Arena's credit-system documentation states that Battle Mode interactions in Text Arena and Code Arena do not count against the daily credit balance.
Other restrictions may apply to individual models or features.
The exact credit allowance and permitted usage can change. Users should consult their current account interface rather than assume that historical limits remain applicable.
Arena's public documentation does not establish a single permanent allowance covering every model and feature.
Do You Need an Account to Use Arena?
An account is useful for preserving conversations and accessing supported features across devices.
According to Arena's account instructions, users can register using either a Google account or an email address.
Email registration requires verification, while Google authentication provides another supported sign-in method.
Some functionality may be available without registration, although particular features can require an authenticated account.
People who previously registered through lmarena.ai do not need to create a different account solely because of the domain change.
Arena's migration guidance explains that existing account holders can sign in through the new website to access their saved conversations.
For users who previously chatted without an account, conversation recovery depends partly on retaining the original browser data.
The official migration instructions warn that manually importing older anonymous chats can replace newer anonymous conversation history. Users should review those instructions before attempting recovery.
How Reliable Are Arena's AI Rankings?
Arena's rankings provide valuable real-world preference data, but they should not be interpreted as definitive measurements of every aspect of AI quality.
Its methodology has several strengths. Anonymous comparisons help reduce brand bias, multiple participants contribute evaluations, and the platform publishes methodological information and research datasets.
Nevertheless, human preference and factual correctness are different measurements.
A confident, well-written answer may receive positive votes even when it contains mistakes. Similarly, an accurate but less engaging response might receive fewer preferences.
Arena has acknowledged this distinction through additional evaluation work, including factuality-oriented rankings that consider the correctness of generated claims.
Leaderboard positions can also change as additional votes arrive, new models are introduced, and evaluation methods evolve.
For important work, users should combine Arena rankings with task-specific testing, independent benchmarks, and direct verification of generated outputs.
Is Arena Safe, and What Happens to Your Conversations?
Arena is an identifiable service operated by Arena Intelligence Inc., but users should understand its data-handling practices before submitting information.
The platform's official FAQ explains that conversations and evaluation feedback may contribute to research, service improvements, and evaluations conducted with AI model providers.
Arena states that data shared publicly or with model providers for these purposes is de-identified.
However, de-identification should not be treated as a guarantee that every submitted detail is harmless to disclose.
Avoid entering passwords, financial information, private customer records, confidential business documents, or other sensitive information.
Users should also remember that Arena connects them with third-party AI technologies. Generated responses may contain inaccuracies or inappropriate content.
For routine model comparisons, the platform provides a structured environment for exploring AI capabilities. For confidential or consequential tasks, its privacy policies and the terms of relevant model providers deserve closer examination.
Who Should Use Arena?
Arena is particularly useful for people who want to compare AI models before selecting one for a specific task.
Writers can examine how models handle explanations and editing requests. Developers can compare coding performance. Designers can explore image-generation capabilities, while researchers can investigate model rankings and evaluation methods.
The platform is less suitable as the sole authority for deciding whether an AI model is accurate, secure, or appropriate for professional deployment.
Its strongest practical value comes from combining two sources of information: public rankings based on user preferences and personal comparisons using realistic prompts.
For anyone searching for lmarena.ai today, the correct destination is Arena's current official website. The original service continues under its updated identity, with broader evaluation categories, evolving usage limits, and continuously updated AI model rankings.