Dimensions overall score 7.0

• The mmBERT-base model achieved the best performance in clinical named entity recognition for Portuguese, with a micro F1-score of 0.76.Source proofresult 95%
  Evidenceverified

  The mmBERT-base model achieved the best performance (micro F1 = 0.76), outperforming all other models.

  Implicationpartial

  This is a direct result stated in the abstract and confirmed in the results section with a specific F1 score.

  Verificationverified

  verified

• Iterative stratification was an effective strategy for improving class balance and overall performance in clinical NER.method 90%
  Evidencepartial

  Iterative stratification improved class balance and overall performance.

  Implicationpartial

  The abstract explicitly states this as a finding, and the results section implies its positive impact.

  Verificationpartial

  partial

• Multilingual BERT models, particularly mmBERT, perform strongly for Portuguese clinical NER and can run locally with limited computational resources.technical 90%
  Evidencepartial

  Multilingual BERT models, particularly mmBERT, perform strongly for Portuguese clinical NER and can run locally with limited computational resources.

  Implicationpartial

  This claim is directly stated in the abstract and supported by the results showing mmBERT's superior performance and the technical advantage of local execution.

  Verificationpartial

  partial

• BERT models generally demonstrated better performance compared to leading LLMs for Portuguese clinical NER.result 85%
  Evidencepartial

  BERT models, in general, have demonstrated better performance when compared to leading LLMs in the industry.

  Implicationpartial

  The results section states this comparison, highlighting BERT's advantage over LLMs in this specific task.

  Verificationpartial

  partial

• The study compared BioBERTpt, BERTimbau, ModernBERT, and mmBERT with LLMs such as GPT-5 and Gemini-2.5.method 95%
  Evidencepartial

  We compared BioBERTpt, BERTimbau, ModernBERT, and mmBERT with LLMs such as GPT-5 and Gemini-2.5

  Implicationpartial

  This is a direct statement of the models included in the comparison, found in both the abstract and methods section.

  Verificationpartial

  partial

• The mmBERT-base model established a new state-of-the-art on the SemClinBr dataset with a micro F1-score of 0.7646.result 95%
  Evidencepartial

  mmBERT-base established a new state-of-the-art on the SemClinBr dataset (micro F1-score = 0.7646).

  Implicationpartial

  This is a specific and verifiable result with a precise F1 score, directly stated in the results section.

  Verificationpartial

  partial

• LLMs, when prompted with increased reasoning effort, showed a modest performance gain but did not surpass every BERT model.result 85%
  Evidencepartial

  The experiments involving increased reasoning effort using the gpt-5 model demonstrate a modest gain in performance that was not able to surpass every BERT model.

  Implicationpartial

  The results section discusses the impact of increased reasoning effort on LLMs, noting the limited performance improvement.

  Verificationpartial

  partial

• The mmBERT-base model achieved the best performance in clinical named entity recognition for Portuguese, with a micro F1-score of 0.76.Source proofresult 95%
  Evidenceverified

  The mmBERT-base model achieved the best performance (micro F1 = 0.76), outperforming all other models.

  Implicationpartial

  The abstract and results section explicitly state this finding and provide the specific F1-score.

  Verificationverified

  verified

• Iterative stratification is an effective strategy for improving class balance and overall performance in clinical NER tasks.method 90%
  Evidencepartial

  Iterative stratification improved class balance and overall performance.

  Implicationpartial

  The abstract and results section mention the exploration and positive impact of iterative stratification.

  Verificationpartial

  partial

• Multilingual BERT models, particularly mmBERT, perform strongly for Portuguese clinical NER and can be run locally with limited computational resources.technical 95%
  Evidencepartial

  Multilingual BERT models, particularly mmBERT, perform strongly for Portuguese clinical NER and can run locally with limited computational resources.

  Implicationpartial

  This is a direct conclusion stated in the abstract, highlighting both performance and technical feasibility.

  Verificationpartial

  partial

• BERT models generally demonstrate better performance compared to leading LLMs for Portuguese clinical NER, with faster inference and local execution capabilities.result 90%
  Evidencepartial

  BERT models, in general, have demonstrated better performance when compared to leading LLMs in the industry. They also had faster inference and ran locally on domestic hardware, enabling experiments with no privacy risks.

  Implicationpartial

  The results section directly compares BERT models to LLMs and highlights these advantages.

  Verificationpartial

  partial

• The study evaluated BioBERTpt, BERTimbau, ModernBERT, and mmBERT models against LLMs such as GPT-5 and Gemini-2.5 for clinical NER in Portuguese.method 100%
  Evidencepartial

  We compared BioBERTpt, BERTimbau, ModernBERT, and mmBERT with LLMs such as GPT-5 and Gemini-2.5, using the public SemClinBr corpus and a private breast-cancer dataset.

  Implicationpartial

  This is a direct statement of the models compared in the abstract and methods section.

  Verificationpartial

  partial

Competitive landscape

A benchmark of modern BERT and LLMs for clinical named entity recognition in Portuguese, demonstrating strong performance with mmBERT and balanced data strategies.