# OmniGuard-7B ![Organization](https://img.shields.io/badge/Org-Independent%20Research-555) ![Pricing](https://img.shields.io/badge/Price-Open%20Weights-2da44e) ![License](https://img.shields.io/badge/License-Research%20Use-d29922) ![Modality](https://img.shields.io/badge/Modality-Text-8957e5) > **Quick answer:** OmniGuard-7B is a 7B open-weights safety classifier from an academic research release (published under an anonymized/pseudonymous Hugging Face organization pending peer review), aimed at general-purpose guardrail classification across multiple harm taxonomies. ## At a Glance **Where it leads:** No individually reported score currently appears in Mistral's public Shieldstral benchmark comparison; included here as a referenced comparison model in the underlying research literature. **Where it lags:** Limited public benchmark disclosure relative to industrially-backed guard models (OpenAI, Alibaba, NVIDIA, Meta, Google). **Best for:** Researchers evaluating academic guardrail baselines alongside industrial guard models. ## What OmniGuard-7B Is OmniGuard-7B is a research-released 7B safety classifier, published via an anonymized Hugging Face organization consistent with double-blind peer review conventions. Its associated paper explores general-purpose safety classification across harm categories, positioning it as an academic baseline within the broader guardrail model literature that includes WildGuard, Aegis, and PolyGuard. ## Specifications | Field | Value | |---|---| | Organization | Independent research (anonymized release) | | Parameters | 7B | | License | Research use | | Modality | Text only | ## Pricing Open weights, free to download for research use. ## Public Benchmark Scores *Scores from Mistral AI's Shieldstral model card, shown for context. Not Benchgen measurements.* | Benchmark | Score | |---|---| | [VLGuard](/benchmarks/independent-research/vlguard) | 88.5% | | [UnsafeBench](/benchmarks/independent-research/unsafebench) | 72.6% | | [LlavaGuard](/benchmarks/tu-darmstadt-hessian-ai/llavaguard) | 71.7% | ## Related Models & Benchmarks - [Shieldstral 1.0](/models/mistral-ai/shieldstral-1-0) · [LlavaGuard-7B](/models/tu-darmstadt-hessian-ai/llavaguard-7b) - [VLGuard](/benchmarks/independent-research/vlguard) · [UnsafeBench](/benchmarks/independent-research/unsafebench) · [LlavaGuard](/benchmarks/tu-darmstadt-hessian-ai/llavaguard) ## Frequently Asked Questions

What is OmniGuard-7B?

A 7B open-weights safety classifier from an academic research release, exploring general-purpose guardrail classification.

Why is the organization "anonymized"?

The model was released under a pseudonymous Hugging Face organization, consistent with double-blind peer review conventions common in academic ML publishing.

Is it open source?

Weights are publicly available for research use.

--- *Last updated 2026-08-12.*

OmniGuard-7B — Evaluations

Evaluations