GLiNER2 large
FastinoOpen weights
The production GLiNER2 checkpoint: entities, classification tasks and structured records in one pass. The best open model on the Metatext suite and the only family that also extracts spans.
- typed-decisions
- 0.486
- Metatext suite
- 0.716
- ECE untuned
- 0.300
- p50 per case
- 627 ms
9/20 over baseline
6/7 over baseline
→ 0.029 held-out fitted
four CPU cores
Specifications
- Checkpoint
- fastino/gliner2-large-v1
- Publisher
- Fastino
- Mechanism
- Schema-conditioned encoder (extraction + classification)
- Backbone
- DeBERTa-v3 (span grid)
- Parameters
- 340M
- License
- Apache 2.0
- Runs as
- open weights, self-hosted
- Library
- gliner2
- Primitives
- Choice, Score, Noul, Extraction
- Pricing
- none — your compute
Metatext decisionsrank 2 of 21
Full leaderboard →Per-question accuracy — default payload, parallel mode, 120 judgments
| Task | Question | Type | Accuracy | Baseline | Beats? |
|---|---|---|---|---|---|
| agent_traces | outcome | choice | 0.571 | 0.357 | ✅ |
| agent_traces | needs_human | noul | 0.643 | 0.714 | ❌ |
| customer_service | department | choice | 0.850 | 0.400 | ✅ |
| customer_service | refund_requested | noul | 0.950 | 0.700 | ✅ |
| customer_service | frustration | score | 0.950 | 0.500 | ✅ |
| security_incidents | severity | choice | 0.438 | 0.312 | ✅ |
| security_incidents | actively_exploited | noul | 0.812 | 0.688 | ✅ |
typed-decisionsrank 3 of 9
Full leaderboard →Per-question accuracy — default payload, parallel mode, 2,000 judgments
| Task | Question | Type | Accuracy | Baseline | Beats? |
|---|---|---|---|---|---|
| td_agent_trace_observability | action | choice | 0.310 | 0.360 | ❌ |
| td_agent_trace_observability | needs_review | noul | 0.450 | 0.550 | ❌ |
| td_agent_trace_observability | outcome | choice | 0.250 | 0.500 | ❌ |
| td_agent_trace_observability | risk | score | 0.460 | 0.340 | ✅ |
| td_agent_trace_observability | urgency | score | 0.300 | 0.370 | ❌ |
| td_customer_service | action | choice | 0.470 | 0.520 | ❌ |
| td_customer_service | category | choice | 0.740 | 0.380 | ✅ |
| td_customer_service | churn_risk | score | 0.660 | 0.470 | ✅ |
| td_customer_service | needs_human | noul | 0.700 | 0.670 | ✅ |
| td_customer_service | urgency | score | 0.610 | 0.420 | ✅ |
| td_invoice_processing | discrepancy_severity | score | 0.160 | 0.470 | ❌ |
| td_invoice_processing | disposition | choice | 0.450 | 0.450 | ❌ |
| td_invoice_processing | duplicate | noul | 0.110 | 0.890 | ❌ |
| td_invoice_processing | matches_order | noul | 0.620 | 0.540 | ✅ |
| td_invoice_processing | urgency | score | 0.320 | 0.510 | ❌ |
| td_security_incidents | credential_compromise | noul | 0.580 | 0.550 | ✅ |
| td_security_incidents | disposition | choice | 0.740 | 0.740 | ❌ |
| td_security_incidents | severity | score | 0.510 | 0.490 | ✅ |
| td_security_incidents | true_positive | noul | 0.730 | 0.730 | ❌ |
| td_security_incidents | urgency | score | 0.540 | 0.500 | ✅ |
Fitted temperature per primitive (fold 0, judged on fold 1): choice 4, score 7, noul 7 — ECE 0.309 → 0.029, accuracy → 0.439.
Frequently asked questions
How does GLiNER2 large compare to Jev?
On typed-decisions GLiNER2 large scores 0.486 against Jev's 0.726, beating the majority baseline on 9 of 20 questions (Jev: 18/20).
Is GLiNER2 large open source?
Yes. fastino/gliner2-large-v1 is released under Apache 2.0 and runs on your own hardware with the gliner2 library.
Can GLiNER2 large extract entities?
Yes. GLiNER2 large has a span head and scores 0.968 micro-F1 on the eight-case extraction task.