AI startup Adaption introduced a new product called AutoScientist in May 2026, designed to help AI models learn specific capabilities quickly through an automated approach to conventional fine-tuning.
AutoScientist works by simultaneously optimizing both training data and the model itself, adapting to whatever capability it is being trained on. The tool builds on Adaption’s existing product, Adaptive Data, which is aimed at building high-quality datasets over time. Together, the two products are intended to create a continuously improving pipeline — from data to finished model.
“What’s super exciting about it is that it co-optimizes both the data and the model, and learns the best way to basically learn any capability,” said Sara Hooker, co-founder and CEO of Adaption and former VP of AI research at Cohere. “It suggests we can finally allow for successful frontier AI trainings outside of these labs.”
Hooker described the company’s broader philosophy as building a fully adaptable stack that can optimize in real time for any given task. The techniques are applicable across a range of fields, though Adaption is particularly focused on speeding up and simplifying the process of training frontier-level AI models.
In its launch materials, Adaption claims AutoScientist has more than doubled win-rates across different models. The company acknowledged that standard benchmarks such as SWE-Bench or ARC-AGI are not applicable to the system, given its task-specific design.
To encourage adoption, Adaption is offering AutoScientist free to use for the first 30 days following its release.
If the tool performs as described, it could reduce the resources and expertise currently required to train or fine-tune frontier AI models — a process that has largely been confined to large, well-funded research labs.
Source: TechCrunch