LMArena went from an academic side project to a $1.7 billion company in barely the time it takes most startups to finish a pitch deck.
The UC Berkeley spinout announced a $150 million Series A on Tuesday at a post-money valuation of $1.7 billion, just four months after launching its first commercial product in September.
The round was led by Felicis and UC Investments, the University of Californiaâs investment arm. Andreessen Horowitz, The House Fund, LDVP, Kleiner Perkins, Lightspeed Venture Partners, and Laude Ventures also joined the deal.
Add in the $100 million seed round it closed in May at a $600 million valuation, and LMArena has now raised $250 million in about seven months.
From research project to billionâdollar business
LMArena started life in 2023 as Chatbot Arena, an open research project from UC Berkeley researchers Anastasios Angelopoulos and Wei-Lin Chiang. It was initially funded through grants and donations.
The idea was simple: crowdsourced leaderboards for AI models.
On its consumer site, a user types a prompt and LMArena forwards it to two different models. The user then picks which response is better. Those head-to-head battles now power public rankings of AI systems across tasks like text, web development, vision, text-to-image, and other capabilities.
According to the company, more than 5 million people use the site each month across 150 countries, generating around 60 million conversations monthly. Model makers quickly became obsessed with climbing those leaderboards.
LMArenaâs comparisons cover a wide range of systems, including versions of OpenAIâs GPT, Google Gemini, Anthropic Claude, and Grok, along with models specialized for image generation, text-to-image, or reasoning.
Turning crowdsourced rankings into revenue
After building a reputation as a neutral testing ground, LMArena began working directly with model vendors. It partnered with companies including OpenAI, Google, and Anthropic to make their flagship models available for the community to evaluate.
That shift triggered some pushback. In April, a group of competitors published a paper alleging that such partnerships helped those model makers "game" LMArenaâs benchmarks. The startup has strongly denied the claim.
The controversy didnât slow down the business.
In September, LMArena publicly launched its commercial product, AI Evaluations, aimed at enterprises, model labs, and developers. Customers pay the company to run structured evaluations of models using its community and infrastructure.
By December â less than four months after launch â that product had reached what LMArena describes as an annualized "consumption rate" of $30 million, effectively its take on annual recurring revenue (ARR).
That growth, layered on top of the platformâs traffic and mindshare in the AI community, was enough to pull in a heavyweight roster of investors for the Series A.
Why this matters
As AI systems proliferate, the question "Which model is actually better for my use case?" is getting harder to answer. Benchmark papers and synthetic tests tell only part of the story.
LMArena is betting that large-scale, human-in-the-loop comparisons will become critical infrastructure for the AI stack â much like browser benchmarks and cloud performance tests did in earlier computing waves.
If the company can maintain trust in its evaluations while scaling a paid enterprise product, its overnight leap from lab project to billion-dollar business may be a preview of how fast the AI tooling ecosystem can move when it sits in the middle of model makers, developers, and end users.
