Claude 3 Haiku: benchmark results
Anthropic's Claude 3 Haiku (March 2024), the smallest and fastest tier of the Claude 3 family, aimed at quick responses to simple queries. Provider: Anthropic. Released 2024-03-13. Access: API.
Aggregate Rating 1495 ± 7, rank #803 of 1656 rated models, from 657 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EIBench | 63.24 | Recall | 100 |
| PECC | 27.67 | Pass@3 | 100 |
| RSPC: A Benchmark for Modeling Stress and Psychiatric Conditions in Digitally Mediated Relationships using Psychiatrist Annotations | 53.8 | Macro-F1 (%) | 100 |
| CureMed-Bench | 73.31 | Logical accuracy (%; share of final answers a GPT-4o judge m | 97.9 |
| CureMed-Bench - Language Consistency | 93.43 | Language consistency (%; share of final answers written in t | 97.9 |
| OpenVLM MMT-Bench - Behavior Anomaly Detection | 75 | Score (%) | 97.8 |
| OpenVLM MMT-Bench - Color Recognition | 90 | Score (%) | 97.8 |
| OpenVLM MMT-Bench - Texture Material Recognition | 85 | Score (%) | 95.4 |
| PRISM-VLM - Hallucination Resistance | 74.5 | Per-axis mean score (%; share of trap questions (absent enti | 95.1 |
| OpenVLM MMT-Bench - CIM | 73.3 | Score (%) | 94.7 |
| OpenVLM MMT-Bench - Point Tracking | 85 | Score (%) | 93.9 |
| OpenVLM MMT-Bench - Astronomical Recognition | 88.9 | Score (%) | 93 |
Interactive version: theaggregate.ai/model?slug=claude-3-haiku · How It Works · Data refreshed daily, snapshot 2026-10-20.