Claude 3 Sonnet: benchmark results
Anthropic's Claude 3 Sonnet, the mid-tier model in the Claude 3 family (alongside Opus and Haiku), balancing cost and capability for large-scale deployments. Provider: Anthropic. Released 2024-03-04. Access: API.
Aggregate Rating 1502 ± 10, rank #775 of 1656 rated models, from 467 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SAD - Identity Leverage | 72.68 | Score (%, higher is better) | 100 |
| SAD - Identity Leverage (Situating Prompt) | 78.46 | Score (%, higher is better) | 100 |
| ShoppingMMLU | 72.58 | Average Score (%) | 100 |
| OpenVLM MMMU - Computer Science | 70 | Accuracy (%) | 89.4 |
| OpenVLM MMT-Bench - Interactive Segmentation | 57.1 | Score (%) | 87.6 |
| OpenVLM MMMU - Finance | 66.7 | Accuracy (%) | 84.3 |
| OpenVLM MMMU - Music | 40 | Accuracy (%) | 84.1 |
| CZ-EVAL - Critical Thinking | 68.89 | Accuracy (%) | 81.8 |
| CZ-EVAL - Klokan (Math) | 44.28 | Accuracy (%) | 81.8 |
| OpenVLM MMT-Bench - Counting By Reasoning | 85 | Score (%) | 81.3 |
| SAD - Mini | 61 | Score (%, higher is better) | 80 |
| Situational Awareness Dataset (SAD) | 47.19 | Score (%, higher is better) | 80 |
Interactive version: theaggregate.ai/model?slug=claude-3-sonnet · How It Works · Data refreshed daily, snapshot 2026-10-20.