Pixtral 12B
Compact 12-billion parameter multimodal vision-language model with a flexible vision encoder that handles arbitrary aspect ratios.
Best for: Edge document parsing, architectural blueprint reading, and affordable multimodal intelligence.
Specifications
- Provider: Mistral AI
- Release date: September 2024
- Context window: 128,000 tokens
- Modalities: Text, Vision, Code
- License: Apache 2.0 Permissive Open Source
- Pricing: Apache 2.0 Open Weights; API $0.15 / MTok
- API available: Yes
Capabilities
- Native arbitrary resolution image ingestion
- Diagram and architectural schematic reading
- OCR and chart comprehension
- Compact 12B size