Adaption's AutoScientist raises open-model medical reasoning accuracy to 75.2%

Adaption says its AutoScientist system raised correct answers on medical long-horizon reasoning from 58.1% to 75.2% for an open-weights model.

The company says the system works on any open-weights starting model and needs no retraining.

Adaption says the improved lower-cost model outperformed GPT-6 Astra and Gemini 4 Argon on long medical records, at up to 15.4x lower cost than those two. These results come from Adaption's own research, and the figures have not been independently verified.