# GLiClass (slim)

> GLiClass is an open-source Python library and family of open-weight zero-shot text classifiers from Knowledgator. It scores every candidate label in one forward pass and runs locally through a pipeline or a Ray Serve endpoint.

- Full: https://www.anchorterminal.com/tools/gliclass.md (~6,200 tokens) · this version ~1,580 tokens · JSON https://www.anchorterminal.com/tools/gliclass.json · canonical https://www.anchorterminal.com/tools/gliclass
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · Updated: 2026-10-09

**D · 49.9/100 · rank #697 of 842 · #9 in Decision models · not agent-ready · confidence medium**


Assessment: An Apache-2.0 classifier that scores a whole label set in one encoder pass on the owner's hardware, with single-label, multi-label, hierarchical and few-shot modes. It returns label scores with no calibration claim, the bundled server has no authentication, and the last three test runs on the main branch, on 24 September 2026, failed.

## Facts

- Kind: Model API · vendor: Knowledgator · category: Decision models · legal entity: Knowledgator Engineering Ltd. · provenance 68/100
- Local only (HTTP): pypi `gliclass`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0 (library and the model weights we checked)
- Probe metrics: not measured yet (probes haven't run)
- Library: `gliclass` 0.1.20 on PyPI (21 July 2026), Python 3.10 or newer, `torch>=2.0.0`, `transformers>=5.0.0`, `scikit-learn`, `numpy>=2.0.0`. Extras `serve` (Ray Serve) and `streaming`
- Models: 31 `knowledgator/gliclass-*` repositories on Hugging Face. v3.0 family of July 2025 (edge 32.7M, modern-base 151M, base 187M, modern-large 399M, large 439M), Instruct v1.0 of February 2026 and Multilang of April 2026
- Licence: Apache-2.0 for the library and for `gliclass-base-v3.0`, per the LICENSE file and the model card
- Question types: Single-label (softmax over the labels) or multi-label (sigmoid on each label), hierarchical label sets as a dict, a task prompt, few-shot examples and retrieval-augmented examples on models trained for them. No ordered rubric type
- Limits: Pipeline defaults of `max_classes=25` and `max_length=1024` tokens for text and labels together, with truncation. `ZeroShotClassificationWithChunkingPipeline` splits longer documents
- Serving: `python -m gliclass.serve` starts Ray Serve on port 8000 at `/gliclass`, one text a request, with dynamic batching. Defaults to CUDA. `GLiClassClient` and an in-process `GLiClassFactory` in Python
- Errors: The handler answers `{"error": "..."}` with 400 for a ValueError and 404 for a KeyError, which includes a request without `labels`. Not documented
- Vendor benchmarks: Base v3.0 card, zero-shot F1 averaged over 14 datasets, 0.6764, and 51.61 examples a second on an A6000 at batch size 1. Vendor figures, not ours
- Tests: 151 pytest functions in 7 files, run on Python 3.10, 3.11 and 3.12 with a ruff job. The last three runs on main failed on 24 September 2026
- Releases: 20 versions on PyPI since 0.1.0 on 2 June 2024. Three GitHub releases with one-line notes (0.1.18, 0.1.19, 0.1.20). No changelog file

- Scores: Reliability 43, Performance pending, Schema & documentation 49, Agent ergonomics 60, Security & auth 38, Payments & pricing 60, Task success pending, Maintenance & community 44, Transparency & trust 64 · total over the 7 assessed categories
- Why: Reliability, Scored on the local-package checklist, since GLiClass is a library and weights the owner runs. · Schema & documentation, Read for a model the owner serves. · Agent ergonomics, Read as an API an agent calls for a decision, as with the other decision models. · Security & auth, Read as software the owner runs. · Payments & pricing, Free Apache-2.0 software with nothing to buy, so 20 + 20 + 20 for pricing, free use and no sign-up under the self-hosted rule. · Maintenance & community, Read for an open-weight model with a library. · Transparency & trust, Apache-2.0 for the library and the weights, with `train.py`, the training datasets named on the card and an arXiv paper (30).
- Sources: 14, open questions: 8, both in the full twin
- Capabilities: inference.decision
- JSON: https://www.anchorterminal.com/api/v1/tools/gliclass.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/gliclass.svg` or a link to https://www.anchorterminal.com/tools/gliclass from a page on knowledgator.com or one of its subdomains, or the README of github.com/Knowledgator/GLiClass, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Pass `--host 127.0.0.1` to `python -m gliclass.serve`, or put the port behind your own gateway. The server checks no credential
2. On a machine without a GPU add `--device cpu --dtype float32 --num-gpus-per-replica 0`. The default configuration expects CUDA
3. Send one text a request to `POST /gliclass`. An array in `texts` is cut to its first item without an error
4. Set `multi_label` to false for one label from a set. The default scores each label independently, so scores do not sum to 1
5. Keep text plus labels under the pipeline's 1,024-token `max_length`, or use `ZeroShotClassificationWithChunkingPipeline`. Longer input is truncated silently

## Connect

```bash
pip install gliclass
```

```bash
pip install "gliclass[serve]"
python -m gliclass.serve --model knowledgator/gliclass-edge-v3.0 --port 8000
curl -X POST http://localhost:8000/gliclass \
  -H "Content-Type: application/json" \
  -d '{"text": "This is a great product.", "labels": ["positive", "negative", "neutral"], "threshold": 0.3, "multi_label": true}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| OpenAI Decisions API | BB | 71.5 | inference.decision | https://www.anchorterminal.com/tools/openai-decisions-api.min.md |
| Decider | B | 69.5 | inference.decision | https://www.anchorterminal.com/tools/decider.min.md |
| Laya | B | 69.2 | inference.decision | https://www.anchorterminal.com/tools/convai-laya.min.md |
| Kev | B | 67.4 | inference.decision | https://www.anchorterminal.com/tools/jaredpalmer-kev.min.md |
| Vela 2.0 | B | 66.5 | inference.decision | https://www.anchorterminal.com/tools/vela.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)