Benchmark

Arena-Hard

LLM-judge dependent, with known style and length bias.

Released
2024-04-19
Categories
human_preference
Openness
unknown
Source
model_reports
Reported scores
3

Reported scores

model_reports

ModelOrganizationReported valueReportedEvidence
DeepSeek-R1DeepSeek92.32025-01-22evidence
DeepSeek-V3DeepSeek85.52024-12-27evidence
Qwen3-235B-A22B (Thinking)Qwen95.62025-05-14evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.