← All companies

Applications / Official source checked

Arena

What Arena does

Arena (formerly LMArena) is a community-powered AI model evaluation platform that collects real-world human preference data from side-by-side “battles” and converts it into live leaderboards and datasets. The consumer product lets people compare frontier AI model responses across multiple modalities (text, code/web development, search, vision, and more) and vote on which response is better; Arena aggregates those votes into rankings intended to reflect how models perform in practice rather than solely on static offline benchmarks. Arena also sells “AI Evaluations” to enterprises, model labs, and developers, using the same community-driven methodology to run evaluation campaigns grounded in human feedback. Its differentiation centers on (1) large-scale, organic human preference signals, (2) transparent and auditable evaluation methods (including open-sourcing key pipeline components and publishing leaderboards/changelog updates), and (3) extending evaluation beyond single-turn chat into agentic and multimodal use cases such as Agent Mode and specialized arenas.

News

Company record

Aug 14, 2026 · Official · ArenaAgent Leaderboard Improvements: Categories & Task Cost (Agent Arena)

Arena updated Agent Arena to include category/task-cost considerations, aiming to make agent leaderboard comparisons more meaningful across different task types and costs.

Jul 14, 2026 · Official · ArenaFactuality in the Arena

Arena launched a leaderboard component that ranks models using a weighted combination of human preference and a factuality audit, initially as a non-default toggle in Text and Search arenas.

Jun 29, 2026 · Official · ArenaArena Reaches $100M in 8 Months

Arena reported hitting a $100M annualized run rate in eight months and shared community scale metrics, including monthly users, conversations, and votes; it also referenced momentum of Agent Mode.

Jun 04, 2026 · Official · ArenaAgent Mode (Agentic evaluation experience)

Arena introduced Agent Mode, designed to let users test multi-step agentic tasks in Arena with autonomous planning and tool use inside a sandbox testing environment.

2026-02-?? · Official · ArenaLMArena is now Arena (rebrand and expanded platform)

Arena announced its rebrand from LMArena to Arena and described opening the new arena.ai experience to a broader multimodal community; it also stated additional scale metrics for monthly users and conversations.

Jan 28, 2026 · Official · ArenaIntroducing AutoEval to the Arena leaderboards

Arena introduced AutoEval scores to provide immediate, calibrated model ratings on real tasks while waiting for human votes to accumulate.

Jan 06, 2026 · Official · PRNewswireLMArena Raises $150 Million to Build the World's Most Trusted AI Evaluation Platform

Arena’s (LMArena’s) Series A announcement: $150M led by Felicis and UC Investments, with participation including Andreessen Horowitz, The House Fund, LDVP, Kleiner Perkins, Lightspeed Venture Partners, and Laude Ventures, at a $1.7B post-money valuation.

May 21, 2025 · Official · PRNewswireLMArena Secures $100M in Seed Funding to Bring Scientific Rigor to AI Reliability

LMArena announced a $100M seed led by a16z and UC Investments with participation from Lightspeed, Laude Ventures, Felicis, Kleiner Perkins, and The House Fund; the release describes a $600M valuation context.

Source map · 5 recurring channels · 16 references

Funding

Latest disclosed valuation$1.7B
Tracked capital$250M2 sourced rounds

What it builds

Arena Leaderboards (multiple arenas)

Arena publishes public leaderboards based on community vote aggregation from battle-style evaluations, across multiple task/modality categories such as Text, Search, Code/Web development, Vision, and additional specialized arenas.

source ↗
Arena Direct Chat / Battle Mode

Arena’s consumer interface lets users submit prompts and compare two anonymous model responses, then vote for the response they prefer; these comparisons feed Arena’s public leaderboards and preference datasets.

source ↗
Agent Mode

Arena introduced Agent Mode to evaluate and test agentic AI in multi-step tasks by allowing the user to switch from single-turn battle mode to an autonomous planning/execution experience inside Arena’s sandbox testing environment.

source ↗
AutoEval

Arena introduced AutoEval scores on its leaderboards to provide immediate, calibrated model ratings on real tasks while waiting for human votes to accumulate.

source ↗
AI Evaluations service

Arena’s paid service provides enterprises, model labs, and developers evaluation campaigns grounded in real-world human feedback.

source ↗

Key metrics

Milestones & partnerships