Verified, then logged.
Sun 22 Nov 2026 · 00:46 KST / 15:46 UTC한국어 | ENSearch /

Breaking01 / AI

Single-pass model gets 46.0% on medical multiple-choice test

The researchers kept the language model fixed and trained only the part that compares questions with answer options.

A research monitor shows a blurred medical multiple-choice question with four options beside a scatter plot linking the question point to the nearest answer point. AI-generated image

Researchers reported on 19 Nov that a model based on Qwen3-8B achieved 46.0% accuracy on a medical multiple-choice test.

The paper used MedMCQA, a collection of medical multiple-choice questions. The model turned each question and its answer options into vectors, then selected the option whose vector was closest to the question’s.

The researchers said accuracy fell to chance when they shuffled the pairings between questions and answer options. The paper gave the chance baseline as 25.0%.

The paper cited earlier results on MedMCQA of 57.6% for Flan-PaLM and 72.3% for Med-PaLM 2. Those models were evaluated under a different procedure.

The paper did not report the separate effect of using several language model layers together.