Adaption touts AutoScientist gains on long medical records without retraining

Adaption announced on Oct. 29 that a lower-cost model using AutoScientist outperformed GPT-6 Astra and Gemini 4 Argon on long medical records. The company claimed costs were up to 15.4 times lower, but the text of its post did not specify which model was used for that maximum cost comparison. It also reported that correct answers rose from 58% to 75% without retraining.

In a quote-post of Adaption’s announcement, EstecioJunior identified the lower-cost model as DeepSeek V4 Flash and the test as Wisedocs’ MLCR. EstecioJunior gave the figures as 58.1% and 75.2%, with the model’s weights unchanged. Those more precise details appear in the quote-post, not in the text of Adaption’s announcement.

sarahookr described the work as a collaboration with @denyslinkov and @Wisedocsai on medical document retrieval. nomotvation called the result “the claim worth checking” and questioned whether it would hold outside Adaption’s setup. Adaption’s post text gives neither a benchmark protocol nor the calculation behind its cost comparison.