Currently listed through these providers:
Model details
Ministral 14B
Ministral 3 14B represents the flagship of the Ministral 3 family, designed as a compact vision-language model that punches above its weight class. Built with 14 billion parameters, this instruct-tuned model achieves performance levels on par with the larger Mistral Small 3.2 24B, making it a compelling option for developers who need strong multimodal capabilities without the hardware overhead. The FP8 quantization allows it to run within 24GB of VRAM, significantly reducing deployment barriers for local and edge environments. This combination of vision understanding and efficient inference makes it particularly well-suited for applications requiring document analysis, visual question answering, and instruction-following on resource-constrained hardware.
The model builds on Mistral's established small language model lineage with instruction tuning applied for chat and task-specific deployments. Its open weights under Apache 2.0 licensing enable both commercial and research use cases, with weights freely accessible on HuggingFace. The vision architecture integrates seamlessly with function calling and structured output capabilities, positioning it as a practical choice for building agents and automated workflows that need to interpret both text and images. The design philosophy prioritizes accessibility—optimizing for diverse hardware including local setups—while maintaining competitive performance against much larger models.
Quick Info
Powered by- Provider
- DevPass (LLM Gateway)
- Model key
- ministral-14b-2512
- Release date
- Dec 2, 2025
- Last updated
- Dec 2, 2025
- Input modalities
- Output modalities
- Capabilities
Cost
- Input token cost
- $0.20
- Output token cost
- $0.20
Limits
- Output tokens
- 8,192 tokens
- Context window
- 262,144 tokens
Transparent token rates
Compare mistral pricing
Rates are shown per one million tokens. Combined means one million input plus one million output tokens.
Latest news about Ministral 14B
No articles yet. Fetch the latest news to show it here.