Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-2344
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-03T00:02:53.975494Z
status: published
lens: g_space
sector: health
headline: Comparative Accuracy and Safety of Large Language Models in Glaucoma Knowledge Assessment
dek: Prcis: Across glaucoma MCQs, GPT-5.2, Gemini 3 Pro, and Claude Opus 4.5 were highly accurate, whereas DeepSeek-V3.2 underperformed; potentially harmful wrong answers were uncommon.
Purpose: To compare the accuracy and potential patient-safety risk of four large language models (LLMs) on glaucoma multiple-choice questions (MCQs).
Methods: DeepSeek-V3.2, GPT-5.2 (via ChatGPT), Gemini 3 Pro, and Claude Opus 4.5 were tested on two American Academy of Ophthalmology datasets: 129 text-only AAO ONE Network glaucoma sel…
gain_title: GPT-5.2, Gemini 3 Pro and Claude Opus 4.5 reached 92.2% to 96.1% accuracy on 129 glaucoma self-assessment questions, with rare harmful errors.
problem_title: (none)
trace_subject: (none)
gain_reading: GPT-5.2, Gemini 3 Pro and Claude Opus 4.5 reached 92.2% to 96.1% accuracy on 129 glaucoma self-assessment questions, with rare harmful errors.
gain_evidence: accuracy was 69.0% for DeepSeek-V3.2, 93.0% for GPT-5.2, 92.2% for Gemini 3 Pro, and 96.1% for Claude Opus 4.5 | GPT-5.2, Gemini 3 Pro and Claude Opus 4.5 achieved high accuracy on text-only glaucoma MCQs
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers tested four large language models on glaucoma questions from American Academy of Ophthalmology materials: 129 self-assessment items and 32 study-course items. Each question was put to each model once, in a fresh session, with a standardized prompt.

Three models scored in the 80s and 90s, while DeepSeek-V3.2 landed in the 60s. The findings suggest strong exam-style knowledge in some models, but multiple-choice performance does not show safe use with real patients, and the authors call for clinician oversight.
limitation: The test used text-only multiple-choice questions, each queried once per model, and the harm-eligibility screen was post hoc, so results may not reflect real patient interactions.
tag: Evidence-backed gain
key_points: On the 129-question AAO ONE set, accuracy was 69.0% for DeepSeek-V3.2, 93.0% for GPT-5.2, 92.2% for Gemini 3 Pro and 96.1% for Claude Opus 4.5. | On the 32-question BCSC set, accuracy was 65.6%, 93.8%, 84.4% and 87.5% respectively; GPT-5.2 beat DeepSeek-V3.2 after correction. | A post hoc screen found 38 of 161 questions (23.6%) could directly produce potentially harmful wrong answers. | Four potentially harmful wrong answers occurred, with no significant difference across models.
rundown: The study used text-only questions from two American Academy of Ophthalmology datasets. Accuracy was compared with Cochran's Q and pairwise exact McNemar tests, adjusted using the Holm method.

Differences among GPT-5.2, Gemini 3 Pro and Claude Opus 4.5 were not significant on the larger set. DeepSeek-V3.2 was significantly less accurate than all three.

Potentially harmful wrong answers were coded using a prespecified rubric. Harmful errors were uncommon but present, which the authors say underscores the need for clinician oversight.
sources:
- peer_reviewed | Journal of Glaucoma | https://doi.org/10.1097/ijg.0000000000002804 | 2026-11-02
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

819529d92e2bd1a137bd9cbc638e90752132cfe39002dab30b5d8cfc05aa5d0a

previous

0000000000000000000000000000000000000000000000000000000000000000