Enterprise AI in Numbers · Capability and cost
The AI model people prefer most in blind comparisons (LMArena text leaderboard)
1,525.7 points
1,525.7 points (Gemini 4 Argon (High)) as of October 22, 2026, according to LMArena, leaderboard dataset, down from 1,533.1 points on September 30, 2026. The rating of the #1 model on the LMArena text leaderboard, built from people's votes between two anonymous answers to the same prompt, at the last board of each month.
What it means for leaders
Preference rankings show which model employees are likely to favour in everyday chat. Treat a change in the lead as a signal to check user satisfaction, not as proof of better accuracy.
What would change the reading: A lead from a new lab, or a large rating gap between the top two.
The trend
Every reading
| Date | Value | Change | Detail | Source |
|---|---|---|---|---|
| 1,525.7 points | −7.4 points | Gemini 4 Argon (High) | Source | |
| 1,533.1 points | +28.4 points | gemini-4-argon-high | Source | |
| 1,504.7 points | −6.9 points | claude-opus-5-max | Source | |
| 1,511.6 points | +11.3 points | claude-opus-5-max | Source | |
| 1,500.3 points | +1.0 points | claude-opus-4-6-thinking | Source | |
| 1,499.3 points | −1.3 points | claude-opus-4-6-thinking | Source | |
| 1,500.6 points | ±0.0 points | claude-opus-4-6-thinking | Source | |
| 1,500.6 points | −1.0 points | claude-opus-4-6-thinking | Source | |
| 1,501.6 points | +20.4 points | claude-opus-4-6 | Source | |
| 1,481.2 points | −3.2 points | gemini-3-pro | Source | |
| 1,484.4 points | −3.2 points | gemini-3-pro | Source | |
| 1,487.6 points | +22.9 points | gemini-3-pro | Source | |
| 1,464.7 points | −2.0 points | gemini-2.5-pro | Source | |
| 1,466.7 points | +0.2 points | gemini-2.5-pro | Source | |
| 1,466.5 points | −7.9 points | gemini-2.5-pro | Source | |
| 1,474.4 points | −3.1 points | gemini-2.5-pro | Source | |
| 1,477.5 points | +68.2 points | gemini-2.5-pro | Source | |
| 1,409.3 points | −8.8 points | o3-2025-04-16 | Source | |
| 1,418.1 points | +12.2 points | o3-2025-04-16 | Source | |
| 1,405.9 points | +43.2 points | chatgpt-4o-latest-20250326 | Source | |
| 1,362.7 points | +5.0 points | deepseek-r1 | Source | |
| 1,357.7 points | −1.0 points | deepseek-r1 | Source | |
| 1,358.7 points | +25.1 points | o1-2024-12-17 | Source | |
| 1,333.6 points | −1.5 points | o1-preview | Source | |
| 1,335.1 points | −4.0 points | o1-preview | Source | |
| 1,339.1 points | +22.9 points | o1-preview | Source | |
| 1,316.2 points | +30.1 points | chatgpt-4o-latest | Source | |
| 1,286.1 points | −0.9 points | gpt-4o-2024-05-13 | Source | |
| 1,287.0 points | +5.4 points | gpt-4o-2024-05-13 | Source | |
| 1,281.6 points | +24.2 points | gpt-4o-2024-05-13 | Source | |
| 1,257.4 points | +3.8 points | gpt-4-turbo-2024-04-09 | Source | |
| 1,253.6 points | ±0.0 points | claude-3-opus-20240229 | Source | |
| 1,253.6 points | +64.2 points | gpt-4-1106-preview | Source | |
| 1,189.4 points | — | gpt-4-0314 | Source |
Definition and method
- What it measures
- The rating of the #1 model on the LMArena text leaderboard, built from people's votes between two anonymous answers to the same prompt, at the last board of each month.
- Population
- Models ranked on the LMArena text leaderboard (overall category)
- Updated
- Monthly (from weekly boards)
- Source
- LMArena, leaderboard dataset · CC BY 4.0
Caveats
- Votes reward answers people like, which isn't the same as answers that are correct.
- Ratings are relative to the models on the board at the time, so values from different years aren't directly comparable.
Cite this
Enterprise AI in Numbers, "The AI model people prefer most in blind comparisons (LMArena text leaderboard)", https://lucianoayres.com/ai-brief/data/human-preference-leader/, retrieved October 25, 2026. Data: LMArena, leaderboard dataset.