ai/ministral3 repository overview

⁠Ministral 3 Instruct 2512

GGUF version by Unsloth

⁠Description

The Ministral 3 family consists of compact, efficient multimodal language models designed for edge deployment and local inference. All three variants—3B, 8B, and 14B—offer strong instruction-following capabilities, vision support, and broad hardware compatibility. Each model is released in GGUF format across multiple quantization levels, enabling flexible trade-offs between performance and resource usage. All variants are post-trained for instruction tasks, making them ideal for:

• Chat-based applications
• Assistants and agents
• On-device inference
• CPU and GPU constrained environments
• Multimodal (vision + text) use cases

⁠Characteristics

⁠Available model variants

¹: VRAM estimated based on model characteristics.

⁠Use this AI model with Docker Model Runner

docker model run ai/ministral3

⁠Use cases

Private AI deployments where advanced capabilities meet practical hardware constraints:

• Private/custom chat and AI assistant deployments in constrained environments
• Advanced local agentic use cases
• Fine-tuning and specialization
• And more... Bringing advanced AI capabilities to most environments.

⁠Links

• Hugging Face (Mistral AI)⁠
• Hugging Face (Unsloth GGUF)⁠
• Unsloth Dynamic 2.0 GGUF⁠