GLiNER2.5-Decide
Fastino's open-weight decision model. A DeBERTa-v3-large encoder that scores label sets supplied at runtime, answers single-label, multi-label, yes/no and ordinal questions in one pass, and extracts spans and relations.
GLiNER2.5-Decide joins classification, extraction and cross-decision rules in one constrained decode; nothing is generated. Fastino reports 60.2% on its own fast-decisions suite of 17 datasets (the launch post says 60.1%) and 38 ms median on a V100, 167 ms on a 48-vCPU CPU. The card says 340M parameters; the safetensors header counts 486M. Two Apache-2.0 siblings share the recipe: GLiNER2.5-Decide-1B (59.6%) and GLiNER2.5-multi-Decide (287M, multilingual, 56.7%). It loads through the gliner2 package rather than the /v1/systemone wire format, and is also served through Fastino's API.
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one calibrated probability
- classify — assigns a category from a fixed taxonomy
- extract — pulls spans or fields out of the input
- route — sends the input to one of several destinations
At a glance
| Parameters | 340M |
| Base model | fastino/gliner2-large-v1 |
| Maker | Fastino Labs |
| Released | 2026-09-24 |
| License | apache-2.0 |
| Reported accuracy | 60.2% |
| Reported latency | 38 ms p50 on V100; 43–47 ms on T4/L4/A100; 167 ms on a 48-vCPU CPU |
Hosted API
Served by Fastino. Get access.
Get the weights
pip install systemonemodels
systemone pull fastino-labs/gliner2-5-decide
The files are served from the maker's Hugging Face repository, fastino/GLiNER2.5-Decide, and verified against the checksums recorded here.