This was created purely as an attempt to turn MedGemma-27B into a thinking and reasoning model. I will share it with everyone, so I would appreciate your feedback.

###Model Source

Merged publicly available Gemma3 models with outstanding scientific reasoning.
google/medgemma-27b-it

###What the model can do

Medical reasoning
Simple scientific reasoning
Simple coding and calculations
Tool calls

Please apply megdgmma thinking-systemprompts.txt when using.

###Future plans

I plan to continue improve this model, Please please support me. Bitcoin address : 34ZuzEUfuVfuEE8PiU1DabiwRzAxGw9JM4
I Uploded model Files.
I adapted the LoRA for MedGemma-27B-it, which was released publicly,to megdgmma3-thinking, and found that the accuracy of reading medical images improved. Therefore, I uploaded the used LoRA (LoRA and Page URL), the LoRA-adapted main model, and its test output as a megdgmma3-thinking-DirectLoRA file. (This file is experimental and may cause performance degradation for non-medical performance.)

##NEWS Amazing! An external user measured the benchmark performance of my model and demonstrated its superiority. (https://arxiv.org/html/2601.10342v1) I am grateful to the authors, Cheng Lin Cheng, Ting Chuan Lin, and Chai Kai Chang, for selecting this my model and quantifying its usefulness.

yaml

models:
  - model: gemma3-thinking(Original blend,Non Uploaded)
    parameters:
      weight: 1.0
  - model: google/medgemma-27b-it
    parameters: 
      weight: 1.618033988749
merge_method: nuslerp
tokenizer_source: google/medgemma-27b-it
dtype: bfloat16

Testament200156/medgemma3-thinking

About this model

Testament200156/
medgemma3-thinking