Adaption
Adaptive AI systems and efficient intelligence · Model developers
What’s disclosed
Adaption works with Nebius on compute infrastructure for its AI research. Co-founder and CTO Sudip Roy described the relationship in a Nebius interview, focusing on reliable access to compute as the company develops systems that adapt to changing tasks and data.
Sources
2 sourcesAdaption shares CTO interview
29 Jul 2026 · Adaption · Web page
The company links its full Nebius interview and describes proportional compute.
Open original sourceCompany context
adaptionlabs.aiAdaption develops an AI platform for developers, businesses and researchers building specialized models and applications. Its Adaptive Data tools improve existing datasets, generate new data and connect sources such as Hugging Face, Kaggle and APIs. Adaptive Intelligence includes AutoScientist, which runs experiments and coordinates data preparation and training to produce custom models users can own and deploy. The company also researches continuous learning during inference and develops Adaptive Interfaces, an initiative exploring how people interact with AI.
Recent updates
6 updatesAdaption launches Visualize for interactive dataset exploration
Adaption released Visualize, the first product from its Adaptive Interfaces programme. Users can ask questions about text and tabular datasets, explore interactive charts, inspect underlying rows and customize charts through natural-language requests.
AdaptionAdaption introduces automated training-data checks in Adaptive Data
Adaption introduced agent-generated checklists that automatically filter training data within its Adaptive Data platform. In company testing across four domains, models trained on checklist-filtered examples achieved relative win-rate improvements of 9.4% to 13.2% over models trained on randomly selected examples.
AdaptionAdaption launches Invent a Dataset for generating custom training data
Invent a Dataset generates structured training examples from a description of the behavior a model should learn. Users can specify dataset size and scope without supplying seed data, a schema or a labeling guide.
Adaption