12,247 multiple choice questions requiring commonsense reasoning about everyday concepts.
Accuracy is the reported evaluation metric for CommonsenseQA. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.
Higher is better
| Rank | Model | Trust | Score | Year | Links | Edit |
|---|---|---|---|---|---|---|
| 01 | gpt-4o | paper | 85.4 | 2025 | Source ↗ | Edit result |
| 02 | Claude 3.5 Sonnet | unverified | 83.2 | 2025 | Source ↗ | Edit result |
| 03 | claude-35-sonnet | paper | 83.2 | 2025 | Source ↗ | Edit result |
| 04 | llama-3-70b | paper | 80.9 | 2025 | Source ↗ | Edit result |
| 05 | Llama 3 70B | unverified | 80.9 | 2025 | Source ↗ | Edit result |
| 06 | BitNet b1.58 2B4T | unverified | 71.58 | 2025 | Paper ↗Code ↗ | Edit result |
| 07 | SmoLM2 (1.7B) | unverified | 43.6 | 2025 | Paper ↗Code ↗ | Edit result |