Bespoke Labs introduced the Nimble model for structured text analysis

Company Bespoke Labs released Nimble — a specialized open-source language model designed for making typed decisions based on text data. The system uses a given scheme of questions to analyze input information and form structured answers.

The development is based on the Qwen3.5-9B architecture, which was fine-tuned on a set of 2,676 carefully selected examples. The use of a specialized dataset allowed the model to surpass the basic version in terms of accuracy in performing tasks that require strict adherence to format and logical consistency.

The tool is focused on scenarios where data extraction or text classification is required according to strict business rules. The model works with custom schemes, which allows it to be integrated into document processing pipelines where predictability of output and compliance with specified data types are critical.

The release is available in the GitHub repository under an open license. The developers note that Nimble is optimized for tasks where general-purpose LLM often make errors in formatting or deviate from given instructions when working with complex decision-making schemes.