GLiNER2.5-Decide

GLiNER2.5-Decide

Classifies English text against any label set you pass in, such as intent, routing, sentiment or urgency, in one pass with no prompt template. A small specialist model from Fastino for operational decisions.

Made by fastino on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: GLiNER2.5-Decide is an English text classification model from Fastino. You hand it a piece of text and a set of labels at call time, and it picks the label, or several labels, that fit. It does not write text or reason, and it needs no prompt template. It is built for operational decisions.

Who it suits: Developers who route tickets, tag messages or screen content and want a small local classifier instead of a large language model.

Industry | Who uses it | How they use it | Example
Customer support | Support engineers | Route incoming messages to the right queue | Refund request versus login problem
Banking | Product teams | Label customer requests by intent | A card dispute or a balance question
Content moderation | Trust and safety teams | Score severity, urgency or spam | A report marked urgent
Operations | IT teams | Sort emails and tickets, or decide on a human handoff | A ticket sent to the right team

If you need open-ended answers or multilingual input, you can skip it, since the card points to a separate multilingual version for that.

Where to use it:

• Python programs on Windows, macOS or Linux, through the gliner2 package
• Support desks, inboxes and routing layers that need a fast label for each message
• Agent pipelines that need a yes or no on whether a task is complete

Use case | Who | What they get back
Intent detection | Support teams | One label from your list, such as refund_request
Multi-label tagging | Analysts | Every label whose score passes the threshold
Sentiment and priority | Operations teams | A label from a scale you define

How to set it up:

• Python and pip
• Room for about 1.9 GB of F32 weights
• The gliner2 package with its local extra

1. Install the gliner2 package with the local extra.
2. Load the model with AutoExtractor, which downloads it from Hugging Face.
3. Call classify_text with your text and your labels.

pip install "gliner2[local]"

from gliner2 import AutoExtractor

model = AutoExtractor.from_pretrained("fastino/GLiNER2.5-Decide")

How to use it: Call classify_text with the text and a dictionary that maps a task name to its labels. A single-label task returns one string, and a multi-label task returns every label above the threshold. One call can score several tasks at once.

model.classify_text(
    "My subscription renewed on April 15 for ¥5,400 after the service was already down. Can I get that charge refunded?",
    {"intent": [
        "order_status", "refund_request", "cancel_subscription", "update_payment",
        "login_problem", "shipping_delay", "bug_report", "speak_to_human", "other",
    ]},
)

Tool | What it does
AutoExtractor | Loads the model from Hugging Face
classify_text | Scores text against the labels you pass in
label set | Any list of labels, including ones with descriptions
threshold | Sets which labels count in a multi-label task

Example: You send a customer message asking for a refund after an outage, along with a list of intent labels. The model returns one label, refund_request, so your system can start the refund workflow right away.

For newcomers and experts: Start with one message and four or five clear labels, and see which one comes back. Write labels as short, distinct names for better results.

The card describes a single forward pass with no generated tokens, which keeps calls fast and local. On its exact-match benchmark across 17 domains, GLiNER2.5-Decide averages 60.2%. The same call can answer a question about a passage, classify a book, or score an ordinal scale.

Pass text and a label set
One forward pass scores the labels
Single-label returns one string
Multi-label returns labels above threshold

Good to know:

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: downloads files from Hugging Face, then runs on your own machine
• Not for: reasoning, explaining or open questions, which the card says it does not do
• Not for: multilingual text, since the benchmark suite is English

Title: Exact-match accuracy on fast-decisions, in percent
GLiNER2.5-Decide: 60.2%
GLiNER2.5-Decide-1B: 59.6%
JevK5: 57.6%
GLiNER2.5-multi-Decide: 56.7%
SemIf: 56.4%
GLiFormer large-v1: 49.0%
Laya Router: 46.6%

On Hugging Face, 13 November 2026

• Task: text classification
• Library: transformers
• Parameters: 486M
• Weights: about 1.9 GB in F32; about 243 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 89,759
• Likes: 457
• License: apache-2.0
• Languages: en
• Based on: fastino/gliner2-large-v1
• Quantized versions: 14
• Fine-tunes: 14
• Tags: transformers, safetensors, gliner2, decision-model, text-classification

Links

• Model: https://huggingface.co/fastino/GLiNER2.5-Decide
• Files: https://huggingface.co/fastino/GLiNER2.5-Decide/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download fastino/GLiNER2.5-Decide

Then follow the usage in its model card. Nothing in this kit runs by itself.