Arena, the company behind the LMArena model leaderboard, has raised $200 million at a $3.1 billion valuation, TechCrunch reports. Alongside the funding, it has added alignment rankings that examine whether models take unauthorized actions, misattribute information or claim to have finished work they never completed.
Lightspeed Venture Partners and Khosla Ventures led the Series B. Arena’s January Series A raised $150 million at a $1.7 billion post-money valuation. The new figure is about 82% higher, close to double in roughly ten months.
The business sells analysis of community feedback
Arena began as a UC Berkeley research project in 2023. Its free platform lets people compare model responses and vote for the better result. Its commercial AI Evaluations service, introduced in September 2025, sells performance analysis based on that feedback to model labs and enterprises.
The company reported $30 million in annualized revenue at its January round and a $100 million annualized run rate in June. Those figures imply growth of about 233%, faster than the increase in valuation. An annualized run rate projects a current revenue pace across a year; it differs from revenue collected over a full year.
Alignment adds a different model-selection question
The new category asks whether models follow human intent. TechCrunch describes three failure types:
- Unauthorized action means taking an action the user did not request.
- False attribution means crediting information to the wrong source.
- Deceptive completion means claiming to have completed a task that remains undone.
TechCrunch calls the alignment leaderboard preliminary and reports that OpenAI models currently lead it. That qualification matters when using the ranking to choose an agent for work.
