v1.78.4-nightly
📦 litellm
✨ 2 features🐛 9 fixes🔧 9 symbols
Summary
This release focuses on numerous bug fixes across UI, pricing, and model integrations, including enabling streaming for GPT-OSS on Bedrock and adding new API guardrails.
✨ New Features
- Added cost tracking for /ocr endpoints.
- Added Guardrails for /v1/messages and /v1/responses API.
🐛 Bug Fixes
- Fixed Key Max Budget Removal Error in UI.
- Reverted fake streaming for GPT-OSS in Bedrock; now supports true streaming.
- Fixed pricing for watsonx model family across various models.
- Corrected Gemini 2.5 Flash Image configuration to not set supports_web_search=true.
- Added support for us-gov prefix for AWS GovCloud Bedrock models.
- Fixed exceptions raised when tags were provided as metadata dictionaries.
- Separated OAuth M2M authentication from UI SSO and handled the Introspection endpoint for Oauth2.
- Added glm-4.6 model to pricing configuration.
- Added missing context to benchmark documentation.
🔧 Affected Symbols
GPT-OSS (Bedrock)watsonx modelsGemini 2.5 Flash/ocr endpoints/v1/messages API/v1/responses APIOAuth M2M authenticationUI SSOIntrospection endpoint