MedQA Demographic Bias (Vals)
Science · 2026-04-16
Vals AI's demographic-framing MedQA evaluation. The headline is the macro-average of six demographic conditions, not a direct run on the standard 1,273-item MedQA test.
Top models (higher is better)
| Model | Score |
|---|---|
| O1 | 96.5 |
| GPT-5.1 | 96.4 |
| Gemini 3.1 Pro Preview | 96.4 |
| GPT-5 | 96.3 |
| GPT-5.4 | 96.1 |
| GPT-5 Mini | 96.1 |
| o3 | 96.1 |
| Gemini 3 Pro Preview | 96.0 |
| O4 Mini | 96.0 |
| Opus 4.5 | 95.9 |
| Gemini 3 Flash Preview | 95.8 |
| Opus 4.6 | 95.4 |
| Qwen3.5 Plus (2026-02-15) | 95.2 |
| o3-mini | 94.8 |
| Sonnet 4.5 | 94.7 |
| Grok 4.20 | 94.5 |
| Kimi K2.5 | 94.4 |
| GLM-5 | 94.3 |
| GPT-5.2 | 94.1 |
| DeepSeek-V3.2 | 93.9 |
| GLM-4.7 | 93.7 |
| Opus 4.1 | 93.6 |
| GPT-5 Nano | 93.3 |
| Gemini 2.5 Pro Experimental 03-25 | 93.1 |
| o1 Preview | 93.0 |
| Opus 4 | 92.9 |
| Sonnet 4 | 92.7 |
| Kimi K2 Thinking | 92.6 |
| MiniMax M2.5 | 92.5 |
| Grok 4 | 92.5 |
| Grok 2 | 92.3 |
| GLM 4.6 | 92.2 |
| Grok 4.1 Fast | 92.1 |
| Grok 4 Fast | 92.1 |
| Sonnet 4.6 | 92.1 |
| Gemini 2.5 Flash Preview (09-2025) | 91.4 |
| gpt-oss-120b | 91.4 |
| GPT-4.1 | 91.2 |
| MiniMax M2.1 | 91.2 |
| Gemini 2.5 Flash Preview 04-17 | 91.0 |