GLiNER2

GLiNER2 is an open-source information extraction model from Fastino AI that unifies named entity recognition, text classification and structured extraction in a single lightweight model. Released under the Apache-2.0 license with close to 2,400 GitHub stars and active development in 2026, it follows the original GLiNER’s insight: extraction should be schema-driven, not retrained per entity type.

Overview

Traditional NER pipelines need a fixed label set and a newly trained model for every change. GLiNER2 instead takes your schema as input: describe the entities, fields or classes you want, and the model finds them in text — zero-shot. The second generation folds what used to be three separate model families into one: entity extraction, classification and hierarchical structured-data extraction from documents. Because the model is small and encoder-based, it runs fast on CPU, which makes it practical for high-volume pipelines where calling a large language model for every document would be too slow or expensive.

Key Features

• One model for NER, text classification and structured extraction
• Schema-based prompting: define what to extract, no retraining
• Zero-shot extraction of custom entity types
• Fast CPU inference suited to high-throughput pipelines
• Hierarchical extraction for nested, document-level structures
• Simple Python API for integration into data workflows

💳 Pricing

GLiNER2 is free and open source under the Apache-2.0 license, including the model weights.

Who Is It For

GLiNER2 fits data engineers and NLP teams processing large document streams — support tickets, contracts, product feeds, compliance text — who need controllable, structured extraction at a fraction of LLM API cost.