Arena Raises $200 Million at $3.1 Billion Valuation as AI Evaluation Expands Into Safety

Arena Intelligence, the company behind the Arena and former LMArena chatbot leaderboard, raised $200 million in Series B funding at a $3.1 billion valuation. Lightspeed Venture Partners and Khosla Ventures led the round, with participation from Salesforce Ventures, 01 Advisors, Dell Technologies Capital and Endeavor, TechCrunch reported. Arena’s own announcement confirmed the financing and said annualized revenue has exceeded $100 million.

The valuation marks a steep increase from the company’s $1.7 billion post-money valuation disclosed in an earlier financing. That round was described in a PR Newswire release as a $150 million raise led by Felicis and UC Investments, while later summaries place Arena’s latest valuation at nearly double the earlier figure within about 10 months. The company began as a UC Berkeley research project in 2023 before becoming a commercial evaluation platform.

Arena built its audience through public, crowdsourced comparisons in which people evaluate model responses side by side. The platform has since developed paid evaluation products for AI labs and enterprise customers, creating a business around measuring model quality in real-world tasks. Arena reported passing $100 million in annualized run-rate revenue in June, roughly 8 months after launching its paid AI Evaluations service, according to a June company summary.

That growth reflects a larger shift in the AI market. Model developers need external evidence about performance across coding, writing, research and agentic workflows, while enterprises need comparisons tied to practical use cases. Public rankings attract attention and participation; paid analytics turn that traffic and evaluation data into revenue. The arrangement also gives Arena influence over how customers, investors and developers interpret which models lead.

The new financing arrives alongside Arena’s Alignment Index, which evaluates more than 2 dozen AI models on behaviors tied to reliability and safety. The tests examine unauthorized actions, misattribution and false claims that a task has been completed. Arena frames the index as an effort to measure agentic systems in the situations where models act on behalf of users rather than simply generate text.

Early rankings put OpenAI’s GPT-6.1 Sol at the top of one version, with Anthropic’s Claude Opus 5.5 second, according to coverage of the launch. Results vary between leaderboard configurations, and another published comparison placed Claude models at different positions. That variation matters because alignment scores depend on test design, task selection and how failures are defined. Arena’s expansion therefore makes evaluation methodology as important as the rankings themselves.

Arena’s rise gives model evaluation a larger role in commercial AI decisions. Labs can use the platform to compare releases against competitors, while enterprise buyers can examine behavior beyond benchmark scores and marketing claims. Developers also gain a public reference point for choosing models for coding, research, customer support and autonomous workflows.

The platform’s position brings a governance challenge alongside its business opportunity. A leaderboard that began as a community research project now sells evaluation services to the companies whose models it measures. That commercial structure doesn’t invalidate the rankings, but it increases the need for transparent test protocols, reproducible results and clear separation between public scores and private assessments. The earlier revenue milestone already raised questions about neutrality as evaluation became a fast-growing business.

The round signals investor confidence in evaluation infrastructure as a distinct AI market. Arena’s backers include venture firms, corporate investors and technology-focused funds, bringing both capital and industry connections to a company positioned between model makers and users. Its revenue growth gives the business a commercial base beyond advertising or consumer subscriptions.

That position also puts Arena in competition with internal testing teams, specialist safety evaluators and benchmark providers. Model companies still control many of their own release evaluations, but external rankings offer a common reference across providers. As models become more capable and more autonomous, demand for tests covering truthfulness, task completion and unauthorized behavior will push evaluation firms toward deeper technical work and stricter standards.

Arena plans to use the Series B financing to expand its evaluation platform and extend measurement across AI safety and real-world agentic capabilities. The Alignment Index gives the company a new product surface, but its credibility will depend on whether rankings remain understandable, repeatable and useful to customers with different risk tolerances.

The next phase will test whether Arena can preserve community trust while scaling a paid infrastructure business. Its $3.1 billion valuation rests on two connected bets: that AI labs and enterprises will keep paying for outside measurement, and that public evaluation will shape model adoption as strongly as raw capability benchmarks. If those bets hold, Arena will become a central intermediary in the market for AI performance and safety.