Skip to content
Longterm Wiki
benchmark-result

Claude Haiku 4.5 on OSWorld: 50.7

Child of OSWorld

Metadata

Source Tablebenchmark_results
Source IDDrtKMMtM7g
ParentOSWorld
Children
CreatedApr 24, 2026, 6:42 PM
UpdatedApr 24, 2026, 6:42 PM
SyncedApr 24, 2026, 6:42 PM

Record Data

idDrtKMMtM7g
benchmarkIdHpb8OjdhT9
modelIdClaude Haiku 4.5(ai-model)
score50.7
unitpercent
date2025-10-15
sourceUrl
notesOSWorld-Verified computer-use benchmark. Outperforms Sonnet 4 (42.2%) and far exceeds Sonnet 3.5 (14%). Reported by Anthropic at release.

Source Check Verdicts

unverifiable95% confidence

Last checked: 4/24/2026

Inline sourcing: unverifiable

Debug info

Thing ID: DrtKMMtM7g

Source Table: benchmark_results

Source ID: DrtKMMtM7g

Parent Thing ID: Hpb8OjdhT9