Spring AI 2.0.0-M4 Release Notes

🎯 Highlights

This release includes 2 new features, 10 bug fixes, 13 other improvements.

If you are using Vertex AI, OCI GenAI, or ZhiPu AI integrations, begin planning migration to alternative model providers such as OpenAI, Azure OpenAI, Anthropic, or other supported providers. These deprecated integrations will be removed in a future major release. #5676

The Vertex AI model integration classes have been deprecated and will be removed in a future release. Users should plan to migrate to alternative model providers. #5676
The ZhiPu AI model integration classes have been deprecated and will be removed in a future release. Users should plan to migrate to alternative model providers. #5676
The OCI GenAI model integration classes have been deprecated and will be removed in a future release. Users should plan to migrate to alternative model providers. #5676

Added capability to use Google Search alongside custom tools in Gemini 3.x models, enhancing the search and tool integration capabilities. #5669
Added support for dynamically disabling native structured output functionality, providing more flexibility in output handling. 019267f

Resolved issue where extraBody configuration was being lost when toolDefinitions were specified in API requests. e65d5de
Corrected regression in AzureOpenAiChatOptions where the stop field initializer was not working properly. 6e8e5fe
Corrected handling of string values for TAG and TEXT filter values in Redis vector store filter expression converter. 32c79b0
Resolved key handling issues in Neo4j vector store filter expression converter. d97da30
Corrected identifier parsing logic in the filter expression text parser to handle edge cases properly. ccc29d1
Enhanced the reliability of media fetching operations in the Bedrock proxy chat model implementation. 75e2bd7
Updated and fixed prompt caching tests to work correctly with the Claude Haiku 4.5 model on Bedrock. #5648
Corrected the API key header configuration when using the OpenAI SDK with Azure OpenAI deployments. 0b142aa
Fixed collection field initialization in super-builders to use null defaults instead of empty collections, preventing unintended behavior #5619
Improved type safety in HeaderUtils by adding explicit type parameters to collection operations a20203a

Updated Google Generative AI SDK dependency to version 1.44.0, bringing latest features and improvements. #5669
Upgraded OpenAI SDK dependency to version 4.28.0 for latest OpenAI API features and fixes. 0ad60de
Upgraded the Anthropic SDK dependency to version 2.17.0 for latest features and improvements #5621

Improved stability and reliability of Oracle and PgVector vector store integration tests. 490a369
Resolved flakiness in Bedrock converse integration test assertions for more reliable test execution. 3c38bb4
Corrected issues in Google GenAI auto-configuration test suite. 1560afc
Resolved issues in Mistral AI integration test suite for improved test reliability. aac56ee
Improved internal implementation of filter evaluation logic in SimpleVectorStore for better maintainability. ba9220b
Improved Maven dependency management by adding missing test scope for json-unit-assertj and leveraging Maven transitivity for test dependencies. 86911c4

Thanks to all contributors who made this release possible: