GLiNER2.5-Decide
Fastino's 340M Apache-2.0 decision classifier that scores caller-defined labels in one forward pass for routing, intent and scoring.
Overview
GLiNER2.5-Decide is a 340M English classification model in Fastino's GLiNER2.5 family, built on a DeBERTa-v3-large-based GLiNER2 model. You pass any label set at call time (intent, routing, sentiment, priority, policy or multi-label tags) and get results in a single forward pass with no prompt template or generated tokens; one call can score several heads at once. It loads with AutoExtractor via the gliner2 library and runs locally on CPU or GPU.
On Fastino's fast-decisions suite of 17 domains, the model card reports 60.2% average exact-match accuracy, ahead of a 1B sibling, JevK5, a multilingual variant, SemIf and Laya Router on that suite. It is explicitly a specialist for operational decisions such as customer and banking intent, ticket and email routing, human handoff, agent completion, moderation, severity, urgency and spam, not a general reasoning model. A multilingual variant exists for non-English input.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which caller-defined label fits this text (e.g. intent, route, sentiment)? | — |
How it works
- Load the model with AutoExtractor
- Pass text and label sets for one or more heads
- Read single or multi-label results
- Route or score in your application
Strengths
- Small and local (CPU or GPU)
- Labels defined at call time
- Multiple heads per call
- Apache-2.0
Caveats
- English only (use the multi variant otherwise)
- Benchmark is the vendor's own suite
Getting started
Install gliner2 and load fastino/GLiNER2.5-Decide with AutoExtractor.At a glance
- Type
- Model
- Status
- production
- Language
- Python
- Runtime
- Transformers / gliner2
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Fastino's Apache-2.0 DeBERTa-v3-large decision classifier: caller-defined tasks and labels get probabilities in one forward pass through `gliner2` on CPU or GPU, for operational classification, routing, and ordinal scoring
AnotiaWang/awesome-decision-models
Frequently asked questions
How big is it?
340M parameters.
How does it compare with Jev?
On Fastino's suite it scores 60.2% versus 57.6% for JevK5, per the model card.
Is it multilingual?
No, use GLiNER2.5-multi-Decide for multilingual input.
Related projects
Japanese System One decision model on ModernBERT-ja (314.6M) returning typed choice, score and bool distributions with no generation.
ProjectbetaPythonOEV: 184M-parameter decision engine that scores options in one pass, reporting 0.7705 on typed-decisions and 22 ms per question.
ProjectresearchPythonOpen multilingual System 1 decision engine built on Laya checkpoints with a router that picks the best checkpoint per request.
ModelbetaPythonMangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.
ProjectproductionPython- blink★ 22
Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.
ServicebetaC - Foq★ 11
Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.
ProjectbetaPython