Pixtral Large (25.02) (Amazon Bedrock)

Pixtral Large 25.02 is Mistral AI's multimodal flagship, delivered through Amazon Bedrock as a fully managed, serverless offering. The catalog describes it as carrying roughly 124 billion parameters, placing it among Mistral's largest releases, and as accepting both text and image inputs while producing text outputs. That combination lets a single call handle workflows that previously needed separate OCR or vision steps piped into a language model, covering document analysis, mixed text-and-image reasoning, code generation, and multilingual content work. As a hosted Bedrock model, it inherits the platform's standard inference posture, with prompts and completions kept inside the customer's Bedrock tenancy rather than exposed to the underlying providers.

The model offers a long 128,000-token context window with a maximum output of the cataloged API limit tokens, giving it room for lengthy document or code inputs alongside image attachments. Its practical sweet spot is mixed-modal tasks where keeping text and visual context in one prompt improves coherence, such as grounded document Q&A, chart and figure interpretation, and code reviews that reference screenshots or diagrams. Teams already standardized on AWS can pick it up alongside other Bedrock models without extra infrastructure to manage, while still benefiting from Mistral's frontier-scale architecture.