Arena nearly doubles its valuation in 10 months: AI rankings company now worth $3.1 billion

Alignment Index adds checks on agent behavior to the usual comparison of responses. Arena studied 90,000 real sessions across 27 models. False completion claims occurred in an average of 10% of sessions, and in nearly half of code debugging sessions.

What Arena checks. The index tracks actions taken without permission, false attribution of results and deceptive claims of task completion. A judge model labeled violations using criteria that researchers refined with human input.

In the preliminary Alignment Index as of Oct 8, models received the following scores:

| Model | Score | | --- | --- | | GPT-6.1 Sol | 87.9 ± 1.5 | | Claude Opus 5.5 | 83.2 ± 2.0 | | Grok 4.7 | 82.7 ± 1.3 |

Funding for comparisons. On Oct 8, Arena announced a $200 million Series B led by Lightspeed Venture Partners and Khosla Ventures. In January, the company raised $150 million at a post-money valuation of $1.7 billion. Investors now value it at $3.1 billion, nearly double in 10 months.

On Jun 29, Arena CEO Anastasios Angelopoulos reported an annualized revenue run rate of $100 million, reached in 8 months. By Oct 8, Agent Arena had logged 7 million sessions less than 5 months after launch. The entire platform had accumulated 350 million sessions and 62 million user votes.

Arena also serves as an environment for coding tasks, alongside its AI model rankings. In instructions dated Aug 24, the company described how to work in Agent Mode:

1. Enable GitHub Connector. 2. Select a repository and branch, then submit a task. 3. Review the diff and preview. 4. Push the changes and create a PR.

Alignment Index is already available on Arena's website, with filters by model and lab.