LLM Reference
activeMedicalSafety

HealthBench

Metric: Length-adjusted score (higher is better)Introduced: 2025

OpenAI benchmark of 5,000 realistic health conversations graded against physician-written rubrics. LLMReference tracks the length-adjusted score when available to reduce verbosity bias.

Models ranked

3

tracked on this benchmark

Score band

57.0 – 55.8

best → lowest tracked

Snapshot trend

—

need ≥2 snapshots

Leaderboard

Tracked models ranked by Length-adjusted score (higher is better).

Compare candidates
#Model variant and provenanceScore
1
GPT-5.6 Sol
Version: length-adjusted; raw 55.6; mean response 1764 charsHarness: Not recordedEvaluator: Not recordedObserved: Jun 26, 2026Confidence: Not recordedSource

Notes: Official GPT-5.6 preview system-card row. Raw score is recorded only in the version label so rankings use the comparable length-adjusted metric.

57.0
2
GPT-5.6 Terra
Version: length-adjusted; raw 58.7; mean response 2285 charsHarness: Not recordedEvaluator: Not recordedObserved: Jun 26, 2026Confidence: Not recordedSource

Notes: Official GPT-5.6 preview system-card row. Raw score is recorded only in the version label so rankings use the comparable length-adjusted metric.

57.0
3
GPT-5.6 Luna
Version: length-adjusted; raw 55.4; mean response 1930 charsHarness: Not recordedEvaluator: Not recordedObserved: Jun 26, 2026Confidence: Not recordedSource

Notes: Official GPT-5.6 preview system-card row. Raw score is recorded only in the version label so rankings use the comparable length-adjusted metric.

55.8

How to read this benchmark

This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.

Trust this score when

  • There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
  • The model list covers the same version family you can actually deploy today.
  • Top candidates overlap with your required routing and feature requirements.

Be cautious when

  • There is only one benchmark snapshot or the dataset appears stale.
  • The benchmark metric direction is opposite of your decision objective.
  • The score difference between options is narrow and likely within implementation variance.

Related benchmarks

Last reviewed: Jun 26, 2026