Medicine
MedHELM MEDEC
MedHELM medec: per-(model, item) medec_error_flag_accuracy in {0,1} (1=correct) on 597 detecting whether a clinical note contains a medical error (error-flag accuracy). 13 models from the MedHELM v4.0.0 release.
597items
13subjects
100%observed
Apache-2.0license
medicinedomain
textmodality
Response matrix
Fit to width. Hover for subject & item; click a cell for details.
Correct (1)Incorrect (0)Unobserved
Scale: 1 = correct · 0 = incorrect
Subjects
- 1openai/o3-mini-2025-01-310.6868
- 2openai/o4-mini-2025-04-160.6667
- 3anthropic/claude-3-7-sonnet-202502190.6281
- 4openai/gpt-5-2025-08-070.598
- 5google/gemini-2.0-flash-0010.5963
- 6deepseek-ai/deepseek-r10.5913
- 7openai/gpt-4o-2024-05-130.5796
- 8openai/gpt-5-mini-2025-08-070.5645
- 9google/gemini-2.5-pro-preview-05-060.5477
- 10openai/gpt-4o-mini-2024-07-180.5327
- 11meta/llama-3.3-70b-instruct0.5293
- 12google/gemini-1.5-pro-0010.5243
- 13anthropic/claude-3-5-sonnet-202410220.5209