langcache-embed-v3-small, Mellum2-12B-A2.5B-Thinking, and LightOnOCR-2-1B models now available on Amazon SageMaker JumpStart

This article announces the availability of three new foundation models on Amazon SageMaker JumpStart: langcache-embed-v3-small for semantic caching, Mellum2-12B-A2.5B-Thinking for code reasoning, and LightOnOCR-2-1B for document OCR.

• langcache-embed-v3-small optimizes semantic caching by mapping queries to dense vectors, reducing redundant LLM calls and accelerating response times
• Mellum2-12B-A2.5B-Thinking uses Mixture-of-Experts architecture for code generation, debugging, and multi-step reasoning with 131,072-token context length
• LightOnOCR-2-1B provides multilingual document-to-text conversion for PDFs and images without brittle OCR pipelines
• Deploy models with a few clicks via SageMaker JumpStart console or Python SDK

These models expand SageMaker JumpStart's foundation model portfolio, enabling customers to deploy specialized AI solutions for caching, coding, and document processing on AWS infrastructure.