Petroc Taylor
Research expert covering global developments in the use of data
Get in touch with us now| Characteristic | 2023 | 2024 | 2025 |
|---|---|---|---|
| GPT-4 | 7.2 | - | - |
| GPT-4 Turbo | 6.9 | - | - |
| GPT-3.5 Turbo | 1.69 | - | - |
| DeepSeek V3 (DeepSeek) | - | 15.86 | - |
| Llama 3.1 405B Standard | - | 9 | - |
| Mistral Large 2 (AWS) | - | 7.45 | - |
| Mistral Large 2 (Azure) | - | 5.34 | - |
| o1 | - | 4.59 | - |
| Claude 3.5 Haiku | - | 4.54 | - |
| Llama 3.1 70B Standard | - | 4.36 | - |
| Llama 3 70B | - | 2.96 | - |
| GPT-4o (May) | - | 2.46 | - |
| Claude 3 Haiku | - | 1.88 | - |
| GPT-4o (Nov) | - | 1.68 | - |
| DeepSeek V3.2 Exp | - | - | 23.24 |
| DeepSeek V3.2 | - | - | 23.13 |
| GPT-5 (high) | - | - | 21.85 |
| o3-pro | - | - | 21.77 |
| GPT-5 mini (high) | - | - | 14.9 |
| GPT-5 (medium) | - | - | 13.08 |
| Grok 4 | - | - | 11.9 |
| GPT-5 (low) | - | - | 8.35 |
| Kimi K2 Thinking | - | - | 7.65 |
| GPT-5 nano (high) | - | - | 7.48 |
| o4-mini (high) | - | - | 6.26 |
| Grok 3 Fast | - | - | 5.57 |
| Claude 4 Opus | - | - | 5.32 |
| GPT-5 mini (medium) | - | - | 5.13 |
The chart is not accessible to screen readers. Please switch to the table view to access the data.
Loading statistic...
Download
Source
Sources
In cooperation with
Stanford University
Release date
April 2026
More information
Region
Worldwide
Survey time period
2025
Supplementary notes
Selected models
Citation formats
