LiteLLM v1.107.0-dev.3 Adds Docker Image Signing and Expands AI Provider/Model Support
The latest LiteLLM release introduces Docker image signature verification, expands OpenAI-compatible provider options, adds new generative models and endpoints, and includes operational improvements relevant to developers deploying and managing AI workloads.
What changed?
LiteLLM v1.107.0-dev.3 introduces several new features and provider integrations. Key among them: - All Docker images for LiteLLM are now signed using cosign, with instructions for verification using either cryptographic commit hashes or release tags. - Support for MiniMax-H3 and H3-Max video generation has been added and is accessible through the /v1/videos endpoint. - The set of OpenAI-compatible providers now includes various Inference APIs. - Together_ai NVIDIA-Nemotron-3.5-Super-VL has been added to the model cost map. - The deepseek-v4.1-flash-maas model has been added for Google Vertex AI. - Amazon Bedrock now supports the jp.amazon.nova-2-lite-v1:0 geo inference profile. - Azure PTU deployments can be managed with per-team PTU spending caps and reporting. - There is now native support for OpenRouter responses in the Rust integration. - Batch processing can store each request and response line in S3 or GCS. - The agent gateway supports Databricks Agent Platform agents. - The proxy can now report when each per-model budget window resets. - MCP now relays authorized modern upstream subscriptions. Numerous fixes and optimizations are also included.
Why does it matter to an everyday developer?
The addition of Docker image signing increases security, ensuring developers can verify image integrity and authenticity before deployment. The expansion of provider and model support means developers have more flexibility and choice for generative AI workloads across multiple clouds and services, with a unifying OpenAI-compatible interface. New operational features, such as batch request storage and per-team PTU governance, assist with auditability, compliance, and team-based cost management. Fixes and improvements help maintain reliability in production environments.
What can the developer do now?
Developers deploying LiteLLM using Docker should start verifying image signatures using cosign, following the provided instructions for either commit hash or release tag verification. Cloud and AI-native app builders can begin experimenting with MiniMax-H3 video generation, Inference API providers, and new supported models through familiar OpenAI API endpoints. Teams leveraging Azure can use the updated cost controls and reporting features. Batch processing pipelines can be updated to utilize S3 or GCS for detailed request/response storage. Overall, developers should review the expanded provider and model integrations for new project opportunities and update their security practices for Docker deployments.