WHBench: leaderboard

Open-ended women's health questions from WHBench (2026): 47 clinician-designed scenarios in ten topics, each built around a typical LLM failure mode such as outdated guidelines, dosage errors or missed health disparities. A Claude Sonnet 4.6 judge grades each answer against four to six independent clinician reference answers with a 23-criterion rubric that penalizes unsafe advice heavily, and the score is the mean normalized rubric score in percent; higher is better.

Metric: Mean normalized rubric score (%), the raw score from -58 to +92 mapped to 0-100 on the 47 expert-written WHBench scenarios in ten women's health topics (fertility, hormonal health, pregnancy, PCOS, contraception and others), three runs per question at temperature 0 with a 4,096-token cap, each response graded by a Claude Sonnet 4.6 judge against 4-6 clinician reference answers and a 23-criterion rubric with asymmetric safety penalties; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 22 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.672.1#113
2Claude Sonnet 4.667.1#173
3GPT-5.466.8#135
4Gemini 3 Flash (Preview)64.7#152
5O363.6#237
6DeepSeek V3.261.3#419
7Grok 360.7#611
8Mistral Large 360.2#832
9Grok 457.9#319
10DeepSeek R152.9#482
11GPT-4.151.8#505
12Grok 3 Mini50#565
13Gemini 2.5 Flash49.5#463
14Claude Opus 449.1#326
15Claude Sonnet 448.1#410

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=whbench · How It Works · Data refreshed daily, snapshot 2026-11-21.