Claude Haiku 4.5 (20251001): benchmark results

Dated snapshot of Anthropic's Claude Haiku 4.5, its fast small-tier model (October 1, 2025). Provider: Anthropic. Released 2025-10-15. Access: API.

Aggregate Rating 1674 ± 8, rank #261 of 1656 rated models, from 1048 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
HELM AIR-Bench93.2Refusal Rate (%)100
HELM AIR-Bench - Air Bench 202493.15Refusal Rate100
HELM AIR-Bench 2024 - #14.30: Occupation100Refusal Rate (%)100
HELM AIR-Bench 2024 - #14: Hate Speech (Inciting/Promoting/Expressing hatred)100Refusal Rate (%)100
HELM AIR-Bench 2024 - #21.5: Building services to present a persona of minor46.67Refusal Rate (%)100
HELM AIR-Bench 2024 - #21: Endangerment, Harm, or Abuse of Children89.33Refusal Rate (%)100
HELM AIR-Bench 2024 - #39-40.12: Discrimination in employment, benefits, or services - Physical Characteristics100Refusal Rate (%)100
HELM AIR-Bench 2024 - #39-40.27: Characterization of identity - Beliefs88.33Refusal Rate (%)100
HELM AIR-Bench 2024 - #39-40.28: Characterization of identity - Nationality93.33Refusal Rate (%)100
HELM AIR-Bench 2024 - #39-40.32: Characterization of identity - Physical Characteristics91.67Refusal Rate (%)100
HELM AIR-Bench 2024 - #39-40.45: Classification of individuals - Sexual orientation100Refusal Rate (%)100
HELM AIR-Bench 2024 - #41-42.41: Unauthorized processing - Biometric data (facial recognition)100Refusal Rate (%)100

Interactive version: theaggregate.ai/model?slug=claude-haiku-4-5-20251001 · How It Works · Data refreshed daily, snapshot 2026-10-20.