Claude Haiku 4.5 (20251001) (Thinking) — benchmark results

Claude Haiku 4.5 (20251001) evaluated with thinking enabled. Provider: Anthropic. Released 2025-10-01. Access: API.

Unified ELO 1654 ± 17, rank #358 of 1776 rated models, from 32 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
UGI - Writing49.08Writing Score88.5
UGI - Natural Intelligence36.76NatInt Score85.8
Vals AI MedScribe85.23Accuracy (%)81.2
Vals AI MGSM92.15Accuracy (%)66.4
Vals AI AIME82.71Accuracy (%)54.7
Vals AI CorpFin v260.61Accuracy (%)53.3
Vals AI LegalBench81.24Accuracy (%)51.2
Vals AI MortgageTax62.16Accuracy (%)48.8
Vals AI Terminal-Bench 2.038.2Accuracy (%)46.2
Vals AI CaseLaw v256.48Accuracy (%)45.4
Vals AI GPQA72.22Accuracy (%)38.9
Vals AI Finance Agent46.93Accuracy (%)38

Interactive version: theaggregate.ai/model?slug=claude-haiku-4-5-20251001-thinking · How the rankings work · Data refreshed daily, snapshot 2026-07-22.