MétaCan
Menu
← Back to cohort
Record W4417474482 · doi:10.2196/77357

Performance of DeepSeek-R1, ChatGPT (GPT-o3-mini), and Gemini 2.0 Flash on German Medical Multiple-Choice Questions: Comparative Evaluation

2025· article· en· W4417474482 on OpenAlexvenueno aff
A. Meyer, Yassin Karay, Andrea U. Steinbicker, Thomas Streichert, Remco Overbeek

Bibliographic record

VenueJMIR Formative Research · 2025
Typearticle
Languageen
FieldMedicine
TopicArtificial Intelligence in Healthcare and Education
Canadian institutionsnot available
Fundersnot available
KeywordsGermanFlash (photography)German government

Abstract

fetched live from OpenAlex

BACKGROUND: Despite the transformative potential of artificial intelligence (AI)-based chatbots in medicine, their implementation is hindered by data privacy and security concerns. DeepSeek offers a conceivable solution through its capability for local offline operations. However, as of 2025, it remains unclear whether DeepSeek can achieve an accuracy comparable to that of conventional, cloud-based AI chatbots. OBJECTIVE: This study aims to evaluate whether DeepSeek, an AI-based chatbot capable of offline operation, achieves answer accuracy on medical multiple-choice questions (MCQs) comparable to that of leading chatbots (ie, ChatGPT and Gemini) on German medical MCQs, thereby assessing its potential as a privacy-preserving alternative for clinical use. METHODS: A total of 200 interdisciplinary MCQs from the German Progress Test Medicine were administered to ChatGPT (GPT-o3-mini), DeepSeek (DeepSeek-R1), and Gemini (Gemini 2.0 Flash). Accuracy was defined as the proportion of correctly solved questions. Overall differences among the 3 models were tested with the Cochran Q test, while pairwise comparisons were conducted using the McNemar test. Subgroup analyses were performed by medical domain (Fisher exact test) and question length (Wilcoxon rank-sum test). An a priori power analysis indicated a minimum sample size of 195 questions. RESULTS: All 3 chatbots surpassed the conventional passing threshold of 60%, with accuracies of 96% (192/200) for DeepSeek, 94% (188/200) for Gemini, and 92.5% (185/200) for ChatGPT. The overall difference among models was not statistically significant (P=.10) nor were pairwise comparisons. However, incorrect responses were significantly associated with longer question length for DeepSeek (P=.049) and ChatGPT (P=.04) but not for Gemini. No significant differences in performance were observed across clinical versus preclinical domains or medical specialties (all P>.05). CONCLUSIONS: Overall, DeepSeek demonstrates outstanding performance on German medical MCQs comparable to the widely used chatbots ChatGPT and Gemini. Similar to ChatGPT, DeepSeek's performance declined with increasing question length, highlighting verbosity as a persistent challenge for large language models. While DeepSeek's offline capability and lower operational costs are advantageous, its safe and reliable application in clinical contexts requires further investigation.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.008
metaresearch head score (Gemma)0.032
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesMetaresearch
Consensus categoriesnone
DomainCandidate signal: Evaluation · Consensus signal: none
Study designCandidate signal: Observational · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.992
Threshold uncertainty score0.043

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0080.032
Meta-epidemiology (narrow)0.0010.000
Meta-epidemiology (broad)0.0010.001
Bibliometrics0.0010.001
Science and technology studies0.0000.001
Scholarly communication0.0010.002
Open science0.0010.002
Research integrity0.0010.001
Insufficient payload (model declined to judge)0.0040.002

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.326
GPT teacher head0.602
Teacher spread0.276 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

Study designObservational
DomainEvaluation
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations4
Published2025
Admission routes1
Has abstractyes

Explore more

Same venueJMIR Formative Research→Same topicArtificial Intelligence in Healthcare and Education→French-language works237,207→