Arena

Happy Monday.

It’s post Superbowl weekend, y’all. While you were watching the Patriots try to remember that they’re playing football, we were watching all the ads.

On that note, if you think the Anthropic vs OpenAI beef is getting a little too personal, this week’s startup might provide a more objective perspective on which LLMs are performing better!

Ethan and Arek 🦄

• Feb. 19 (SF): Our partners at Lightfield are hosting a live masterclass on founder sales. If you’re a founder or early-stage GTM operator, this one is probably for you.

• Feb. 24 (SF): We’re hosting a seed+ founders dinner with Fidelity Private Shares and Fenwick. RSVPs open next Monday.

• Feb. 26 (SF): We’re hosting a founder fireside with a surprise guest, in partnership with Fidelity Private Shares and Intercom. RSVPs (and the guest announcement) in next Monday’s newsletter. 👀 

Arena (formerly LMArena) is an online platform that allows consumers to view and compare the performance of frontier AI models. With an interface familiar to anyone who has used an AI chatbot like ChatGPT, it allows consumers to chat with the best AI chatbot, including ones they don’t actively have a subscription to. Users can use “Max,” a model router that dynamically directs queries to the most appropriate AI model. The company additionally provides users with two responses they can choose between, essentially crowdsourcing the data to rank models. These analytics can subsequently be provided to AI labs to refine and align their models.

Check it out: arena.ai

Arena sells to enterprises like AI labs looking to improve model performance.

• Raised $150 million Series A from a16z, Kleiner Perkins, Lightspeed Venture Partners, Felicis, Laude Ventures, LDV Partners, The House Fund, and US Investments

• Grew community by over 25x with adoption by AI labs

• Boasts over 50 million votes and 400 model evaluations

• Analytics Engineer

• Data Scientist

• Engineering Manager

• Machine Learning Engineer

• Machine Learning Scientist

• Machine Learning Scientist - Open Source Lead

• Senior Infrastructure Engineer

• Senior Product Security Engineer

• Senior Software Engineer, Data Infrastructure

• Senior Software Engineer, Full Stack

• Senior Software Engineer, ML Infrastructure

• Technical Recruiter

• Enterprise IT Engineer

Arena is founded by a true powerhouse of a team.

Anastasios Angelopoulos, Arena’s CEO, formerly worked at Google DeepMind. He received his bachelor’s in electrical engineering from Stanford and his PhD in machine learning and computer vision from Berkeley.

Arena’s CTO, Wei-Lin Chiang, also has impressive accolades, including internships at Amazon, Google, Alibaba, and Microsoft. He earned both a bachelor’s and master’s from National Taiwan University and is currently pursuing his PhD in EECS at Berkeley.

Last, but definitely not least, is Ian Stoica, a serial founder. He founded Databricks and Anyscale, additionally serving as the CEO for the former. He graduated with his PhD from Carnegie Mellon.

What began as a research project is now a trusted resource on AI model performance worth $1.7 billion.

OpenAI’s ChatGPT was a first of its kind AI chatbot when it was released in 2022. Each successive feature, like image generation and internet search, were quick hits with the public, and were swiftly adopted by competitors. Those companies even introduced features of their own. 

Today, however, the common subset of features among LLMs has grown as the frontier labs have shifted their focus toward more urgent competition in fields like agents. Despite this, chatbots remain the most popular way of interacting with AI, and we anticipate this to be the case even as trendy applications like agents see increased adoption.

But while the feature set of these LLMs become interchangeable, their performance does not. Just this past Friday saw the release of two new models, Opus 4.6 and GPT-5.3-Codex. While OpenAI’s model is currently not available through a chatbot interface, the competing releases more importantly showcase the ongoing struggle to claim the top spot for performance among a number of categories.

This raises a number of questions. On the enterprise side, how do these companies measure the performance of their models? Second, how do they compare model performance with those of competitors? And third, on the consumer side, how do users of AI chatbots choose the best model for a task when models are announced so frequently (and with confusing names like GPT-5.3-Codex, too)?

Each of these is a customer pain point addressed by Arena. Arena removes the need for users to pick a model for a task, at a price that beats a standard ChatGPT Plus subscription, while frontier labs have a method of gauging their models from a source that is as close to their ICP as possible. 

The team, too, is nothing short of impressive, with an equally outstanding list of investors to complement it. Altogether, Arena is a strong team building a useful product with a clever model that provides both parties an improved experience, and it’s one we only see growing in the short and long term. In our view, that’s a winning combination.

• LMArena Raises $150M in Series A Funding [FinSMEs]

• LMArena lands $1.7B valuation four months after launching its product [TechCrunch]

• Introducing Max [Arena blog]

• Fueling the World’s Most Trusted AI Evaluation Platform [Arena blog]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.