The company behind LMArena, the leaderboard that’s become the de facto scoreboard for AI model performance, just raised $200 million at a $3.1 billion valuation. That’s nearly double what the company was worth in December, led by Lightspeed and Khosla Ventures.

Arena started as a research project that let people compare AI models head-to-head in blind tests. It became required reading for anyone tracking the industry. When a new model from OpenAI, Anthropic, or Google drops, the first question isn’t what the company says about it. It’s how it performs on Arena.

Now the company is expanding beyond performance metrics into alignment issues. The new focus: measuring whether models lie, refuse appropriate requests, or comply with harmful ones. That’s a harder problem than benchmarking speed or accuracy, but it’s the one that matters more as these models get deployed at scale.

The valuation jump makes sense when you look at what’s happening in AI evaluation. Companies are spending billions to train models, but there’s no good way to measure what you’re getting. Arena solved that for performance. If they can do the same for safety and alignment, they’ll be even more valuable.

OpenAI’s revenue problem

Meanwhile, OpenAI’s revenue picture looks messier than previously reported. New numbers suggest the company’s annualized revenue is around $50 billion, not the $70 billion figure that had been circulating. That’s a $20 billion gap.

The discrepancy matters because OpenAI’s valuation and its ability to raise capital depend on growth projections. Investors put money into the company based on a certain revenue trajectory. If that trajectory is off by nearly 30%, it changes the math on everything else.

This doesn’t mean OpenAI is failing. $50 billion in annualized revenue would still be extraordinary growth for a company that barely existed as a commercial entity three years ago. But the gap between expectation and reality is the kind of thing that makes investors nervous, especially when you’re talking about the sums of money required to keep training frontier models.

The teen founder and the $500M China round

Two other funding rounds tell different stories about where AI investment is flowing.

Zach Yadegari, the 19-year-old who co-founded Cal AI, raised $10 million for his new personal AI agent startup. He’s competing with Instinct, Muse, and Bee in a space that’s gotten crowded fast. The bet is that personal AI agents will be the next consumer platform, but nobody’s proven that yet. Yadegari has one successful consumer AI app already. Whether that translates to a second hit in a different category is the open question.

In China, Manus raised over $500 million in its first funding round since splitting from Meta. Boyu Capital and IDG Capital led, with Tencent and other existing shareholders participating. The round shows Chinese AI companies can still raise significant capital despite US export restrictions and the broader tech tension between the two countries. It also shows that Meta’s former China partnerships are continuing under new structures, even if the formal relationships have ended.

Google ships an agentic Gemini

On the product side, Google launched agentic AI features for Gemini, starting with business customers. The system can plan and execute tasks across multiple apps, delegate work to specialized sub-agents, and use different AI models depending on the task. It even gets its own workplace identity with an email address.

This is Google’s answer to the agent wave that’s been building since early 2025. The enterprise-first approach makes sense. Businesses will pay for agents that actually work, and they’re more forgiving of the rough edges that consumer users won’t tolerate.

Google also released AI Edge Foresight, a local-first meeting note-taker that competes directly with Granola. It runs on-device, transcribes conversations, generates notes, and answers questions without sending data to the cloud. The privacy angle is the selling point, especially for companies that can’t or won’t send meeting data to external servers.

The broader pattern: AI is professionalizing fast. Arena’s valuation reflects the market’s need for credible measurement. The OpenAI revenue gap shows what happens when expectations run ahead of reality. And Google’s enterprise focus shows where the near-term money actually is. Not in viral consumer apps, but in tools that make businesses more efficient.

The teen founder raising $10 million is the exception, not the rule. Most of the capital is flowing to companies solving hard infrastructure problems or serving customers who’ll pay for solutions that work.