Which AI Labs Get the Most Chatbot Arena Votes?
OpenAI models have drawn the most Arena votes of any lab, yet hold none of the top 20 seats. Votes measure presence over time, not today's rank.
The signal
- LaunchesClaude Opus 5.5 and GPT-6 Sol / GPT-6 Luna were both announced on 22 Sept 2026.
- Arena25 Sept 2026: claude-opus-5.5-high is #1 on 2,307 votes, a ±12-point interval. #2 claude-fable-5.1-max has 9,942 votes.
- Arenagpt-6-sol-max is #153, gpt-6-luna-max is #158, each on fewer than 5k votes.
OpenAI's models hold 17.9% of all 12.3M Arena votes and none of the top 20 seats. Anthropic's hold 17.0% of votes and 9 seats.
Left: each lab's share of all votes cast for the models on the board. Right: share of the top 20 places. Arena text board (overall), 25 Sept 2026. Both in %.
Vote counts are cumulative per model since it joined the board, so labs with many and older models collect more. Arena decides which models to show, so vote share also reflects its sampling. Top-20 seats count every listed model, including effort variants of one model. Two models with a blank organisation field are in "Other labs".
Show the data (6 rows)
| Lab | Models rated | Votes | Share of votes | Top-20 seats |
|---|---|---|---|---|
| OpenAI | 47 | 2197859 | 17.9% | 0 |
| Anthropic | 32 | 2091261 | 17.0% | 9 |
| 40 | 1642288 | 13.3% | 6 | |
| Alibaba (Qwen) | 42 | 1033573 | 8.4% | 1 |
| Meta | 22 | 755638 | 6.1% | 3 |
| Other labs (47) | 226 | 4589233 | 37.3% | 1 |
The Mode read
Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol and Luna were announced on the same day, 22 Sept 2026. 3 days later Opus 5.5 is #1 on the Arena board after just 2,307 votes. Every Arena rank rests on blind votes, so we asked: which labs collect the most of them?
The 25 Sept 2026 board holds 12.3 million votes across 409 models. OpenAI's 47 models hold 17.9%, Anthropic's 32 hold 17.0% and Google's 40 hold 13.3%: 48% between three labs. The top 20 seats look different. Anthropic has 9, Google 6 and OpenAI none. The most-voted model of all is claude-3-opus-20240229, from 2024, with 194,909 votes.
Votes follow presence, not rank. A lab that ships often and stays on the board keeps getting compared, long after its models leave the top. That is a small version of how distribution works in any market: being in the comparison set is the step before being chosen. A startup that drops out of the conversation between launches has to buy its way back in.
Votes follow presence, not rank. Being compared comes before being chosen.
What we would back: Teams that ship in public on a steady cadence and stay in every buyer's shortlist between launches.