Amazon Bedrock adds 18 fully managed open weight models, including the new Mistral Large 3 and Ministral 3 models | AWS News Blog (opens in new tab)

Amazon Bedrock has significantly expanded its generative AI offerings by adding 18 new fully managed open-weight models from providers including Google, Mistral AI, NVIDIA, and OpenAI. This update brings the platform's total to nearly 100 serverless models, allowing developers to leverage a broad spectrum of specialized capabilities through a single, unified API. By providing access to these high-performing models without requiring infrastructure changes, AWS enables organizations to rapidly evaluate and deploy the most cost-effective and capable tools for their specific workloads.

Specialized Mistral AI Releases

The launch features four new models from Mistral AI, headlined by Mistral Large 3 and the edge-optimized Ministral series.

• Mistral Large 3: Optimized for long-context tasks, multimodal reasoning, and instruction reliability, making it suitable for complex coding assistance and multilingual enterprise knowledge work.
• Ministral 3 (3B, 8B, and 14B): These models are specifically designed for edge-optimized deployments on a single GPU.
• Use Cases: While the 3B model excels at real-time translation and data extraction on low-resource devices, the 14B version is built for advanced local agentic workflows where privacy and hardware constraints are primary concerns.

Broadened Model Provider Portfolio

Beyond the Mistral updates, AWS has integrated several other open-weight options to address diverse industry requirements ranging from mobile applications to global scaling.

• Google Gemma 3 4B: An efficient multimodal model designed to run locally on laptops, supporting on-device AI and multilingual processing.
• Global Provider Support: The expansion includes models from MiniMax AI, Moonshot AI, NVIDIA, OpenAI, and Qwen, ensuring a competitive variety of reasoning and processing capabilities.
• Multimodal Capabilities: Many of the new additions support vision-based tasks, such as image captioning and document understanding, alongside traditional text-based functions.

Streamlined AI Development and Integration

The primary technical advantage of this update is the ability to swap between diverse models using the Amazon Bedrock unified API.

• Infrastructure Consistency: Developers can switch to newer, more efficient models without rewriting application code or managing underlying servers.
• Evaluation and Deployment: The serverless architecture allows for immediate testing of different model weights (such as moving from 3B to 14B) to find the optimal balance between performance and latency.
• Enterprise Tooling: These models integrate with existing Bedrock features, allowing for simplified agentic workflows and tool-use implementations.

To take full advantage of these updates, developers should utilize the Bedrock console to experiment with the new Mistral and Gemma models for edge and multimodal use cases. The unified API structure makes it practical to run A/B tests between these open-weight models and established industry favorites to optimize for specific cost and performance targets.