GLiNER2.5 Released: Replaces Span Enumeration with Boundary Prediction for Long-Context Extraction
bclavie · x · 2026-08-25
GLiNER2.5 has been released, marking the most significant upgrade to the GLiNER architecture to date. The new architecture replaces span enumeration with boundary prediction, directly scoring entity start and end positions. This removes limits on entity length and allows inference to scale linearly with document length.
The update introduces five new capabilities:
- Long-context extraction and classification over full documents
- Unlimited span lengths
- Joint information extraction
- Constrained classification
- Span attributes
The model achieves higher average F1 scores across 16 benchmarks compared to GLiNER2, with a 24.75-point gain on XNLI. Three variants are released on Hugging Face under Apache 2.0: gliner2.5-base-v1 (0.2B), gliner2.5-multi-v1 (0.3B), and gliner2.5-small-v1 (74M). Training data was generated by the Fastino Data Agent, covering task combinations, document formats, and languages not found in public datasets.
Related event: GLiNER 2.5 Released with Boundary Prediction for Full-Document Extraction(3 posts)→
More from Models
- Nathan Lambert slams Meta's Muse Open License as 'trash' — natolambert · 2026-10-16
- Google launches Gemini 4 Argon to Ultra and Pro subscribers, touting physics and 3D skills — ShikharMurty · 2026-10-16
- Google gives X users early access to test Argon, not just companies — lyraxana · 2026-10-16
- Grok Voice 3.0 Tops Design Arena Speech Leaderboard, Up 7.4 Points — XFreeze · 2026-10-16
- Robotics LLMs tested on 16 tasks: GPT-6 Astra hits 17.2% success vs Fable's 9.1% — pabbeel · 2026-10-16
- Meta open-sources Muse Spark: top US open model, scores just below DeepSeek — bindureddy · 2026-10-16