18 New Open‑Source Models on Amazon Bedrock—Switch Without Code Changes

At re:Invent 2025 Amazon Web Services announced that Amazon Bedrock adds 18 fully managed open‑source models from Google, MiniMax AI, Mistral AI, Moonshot AI, NVIDIA, OpenAI and Qwen, raising the catalog to almost 100 serverless models.

Mistral Large 3 is optimized for long‑context, multimodal and instruction reliability, excelling in long‑document understanding, agent and tool workflows, enterprise knowledge processing, code assistance, mathematics, multilingual analysis, and visual multimodal reasoning.

Ministral 3 3B is the smallest model in the series, edge‑optimized for single‑GPU deployment, and delivers strong language and vision capabilities for image captioning, text classification, real‑time translation, data extraction, short‑form content generation, and lightweight real‑time applications on resource‑constrained devices.

Ministral 3 8B balances high performance with minimal system resource usage, suitable for constrained environments that require dialogue interaction, image and document description, and specialized agent scenarios, while providing stable performance for local or embedded systems.

Ministral 3 14B offers top‑tier text and vision performance within the series, targeting advanced local agent use cases and private AI deployments where hardware is limited but advanced functionality is required.

All models are reachable through a single Bedrock API, allowing developers to evaluate, switch, and adopt new models without rewriting application code or altering underlying infrastructure. Users can test prompts in the Bedrock Playground or integrate via any AWS SDK using the InvokeModel and Converse APIs, and combine the models with any Bedrock‑compatible agent framework, including AgentCore and Strands Agents.

When deploying publicly available models, customers should verify data‑privacy requirements, monitor output bias, and continuously ensure AI compliance. Amazon Bedrock’s enterprise‑grade security features and Guardrails let users configure responsible AI policies, while the Bedrock model evaluation tool assists in benchmarking and selecting the best model for specific business scenarios.

For detailed usage, refer to the Amazon Bedrock User Guide, especially the code examples for SDK development, and consult the region‑specific model availability lists and pricing pages linked in the article.