guard
Shieldstral 1.0 3B
Mistral AI's guard model that takes a policy at inference as a natural-language yes/no query (no fixed taxonomy), for text and image. HF 5 Aug 2026.
Levelmodel
Classguard
Parameters3B
Contextnot in the notes
Released28 Jul 2026
Opennessopen weights permissive
Runs onnot in the notes
Statusactive
Benchmark Results
vendor-reported
HarmBenchsafety
vendor-reported
99.4
050100
Score in F1. Higher is better.
PolyGuardsafety
vendor-reported
84.6
050100
Score in F1. Higher is better.
VLGuardsafety
vendor-reported
97.7
050100
Score in F1. Higher is better.
WildGuardTestsafety
promptvendor-reported
88.1
050100
Score in F1. Higher is better.
XSTestsafety
refusalvendor-reported
94.6
050100
Score in F1. Higher is better.
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).
Lineage
finetunesuccessor
Ministral-3-3B-Base-2512 is the parent (finetune)high confidencemodel cardWhere this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 3 models. Confidence and evidence are the notes' own.
Identity
- Developer
- Mistral AI
- Family
- not in the notes
- Released
- 28 Jul 2026 sourcesnotes
- Weights published
- –
- Record checked
- 10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.
Architecture and Size
- Architecture
- not in the notes
- Total parameters
- 3B sources
- Active parameters
- –
- Checkpoint parameters
- –
- Note
- 4B listed
- Context length
- not in the notes
- Input
- text, image
- Output
- text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).
Openness and Access
- Openness
- open weights permissive
- Licence
- Apache License 2.0 sources
- Access routes
- open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.
Deployment
- Runs on
- not in the notes
- Footprint
- not in the notes
- Runtimes
- not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.
Indonesia
- Indonesian support
- not stated
- Regional languages
- not in the notes
- Jakarta hosting
- unknown
- Languages
- 12 languages; no Indonesian
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.
Industry Use Cases
No industry use case links this model.
Links
DeveloperMistral AIstartup, France
Mistral AIstartup, FranceModelShieldstral 1.0 3BMistral AI, guard
Shieldstral 1.0 3BMistral AI, guardR&D TopicsGuardrails and Safety Classifiers
Use Casesnone recorded
Who Runs Itnone recorded
Organisations 1
Sources
- huggingface.co/mistralai/Shieldstral-1.0-3B https://huggingface.co/mistralai/Shieldstral-1.0-3B
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Shieldstral 1.0 3B", https://aurion.semai.id/models/mistral-ai-shieldstral-1-0-3b/, records of October 2026.