Claude Haiku 4.5 (Thinking): benchmark results
Claude Haiku 4.5 evaluated with thinking enabled. Provider: Anthropic. Released 2025-10-15. Access: API.
Aggregate Rating 1605 ± 8, rank #697 of 2168 rated models, from 313 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - Toxicity - Fairness | 100 | Criteria met (%) | 98.6 |
| Nejumi 4 - Toxicity - Violation Categories | 57.94 | Criteria met (%) | 97.2 |
| BenchTable - Tech | 85.7 | Weighted Score (%) | 96.4 |
| AA-Omniscience Hallucination Rate | 27.3 | Hallucination Rate (%) | 94.3 |
| ReLE - Reasoning - BBH | 84.3 | Accuracy (%) | 88.5 |
| Nejumi 4 - HalluLens | 99 | Hallucination resistance (%) | 88.1 |
| MM-JudgeBias - Image Dominance | 0.53 | Bias-Deviation (0 to 1) on MM-JudgeBias (about 200 image-que | 86.2 |
| Nejumi 4 - Toxicity - Prohibited Acts | 96.88 | Criteria met (%) | 85.3 |
| Nejumi 4 - jaster (2-shot) - JaNLI | 100 | Exact match (%) | 85 |
| OpenCompass Knowledge - Humanities - English (CompassBench 2510) | 87 | Score (%) | 84.6 |
| OpenCompass LLM - Agent (CompassBench 2510) | 63.4 | Score (%) | 84.6 |
| Nejumi 4 - Toxicity - Social Norms | 98.02 | Criteria met (%) | 84.3 |
Interactive version: theaggregate.ai/model?slug=claude-haiku-4-5-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.