Benchmark Registry

HealthBench

HealthBench results reported for 5 models across 3 versions. Updated 2026-10-02.

Report an issue

Versions

3 versions · Metric: HealthBench Score

VersionReleasedModelsResults
HealthBenchLatestMay 12, 202546
HealthBench HealthBench ConsensusMay 12, 202535
HealthBench HealthBench HardMay 12, 202546

Recently reported results: HealthBench

Latest version: HealthBench

ModelScoreSource
Qwen3.8-Max60.2%huggingface.coHealthBench ScoreReported Model cardQwenSelf-reportedQwen TeamRecord
gpt-oss-120b (medium)55.9%deploymentsafety.openai.comHealthBench ScoreReported Model cardOpenAISelf-reportedRecord
gpt-oss-120b (high)57.6%deploymentsafety.openai.comHealthBench ScoreReported Model cardOpenAISelf-reportedRecord
gpt-oss-120b (low)53.0%deploymentsafety.openai.comHealthBench ScoreReported Model cardOpenAISelf-reportedRecord
OpenAI o3 (high)59.8%deploymentsafety.openai.comHealthBench ScoreReported Model cardOpenAISelf-reportedRecord