Change8

v1.82.6.rc.1

📦 litellmView on GitHub →
10 features🐛 27 fixes🔧 13 symbols

Summary

This release introduces new features like the sagemaker_nova provider and various UI enhancements, alongside numerous bug fixes across providers (Anthropic, Gemini, Vertex AI) and infrastructure improvements, including security updates and better logging redaction.

Migration Steps

  1. If you rely on Prisma migrations failing on pre-existing instances, note that this behavior is now opt-in via the --enforce_prisma_migration_check flag.

✨ New Features

  • Added opt-out flag for default reasoning summary in Anthropic provider.
  • Added sagemaker_nova provider for Amazon Nova models on SageMaker.
  • Added endpoints for creating characters and other new video endpoints.
  • Added Akto Guardrails integration to LiteLLM.
  • Added continuous performance benchmarks via CodSpeed.
  • UI: Added Organization Dropdown to Create/Edit Key.
  • UI: Added functionality for Internal Users to Add/Remove Team Membership.
  • UI: Modernized Default Team Settings page and fixed defaults application.
  • UI: Added auto-pagination for daily spend data in Usage section.
  • UI: Added ability to Disable Custom Virtual Key Values via UI Setting.

🐛 Bug Fixes

  • Preserved thinking.summary when routing Anthropic responses to OpenAI Responses API.
  • Resolved image token undercounting in Gemini usage metadata.
  • Ensured HTTP/SDK clients are not closed on LLMClientCache eviction.
  • Fixed Tag list endpoint 500 error caused by invalid Prisma group_by kwargs.
  • Fixed Team Admin 403 error on /user/filter/ui when Org Scope Flag is present.
  • Redacted secrets from proxy log output.
  • Fixed prisma migrate deploy failures on pre-existing instances.
  • Fixed hanging CI tests in custom_httpx test_http_handler.
  • Fixed LiteLLM redaction filter logging v2 issues.
  • Fixed Vertex AI Batch Output File Download failing with 500 error.
  • Preserved custom attributes on the final stream chunk.
  • Fixed alignment of DefaultInternalUserParams Pydantic default with runtime fallback.
  • Fixed privilege escalation issues on /key/block, /key/unblock, and /key/update max_budget.
  • Fixed UI Logs showing stale data when filter results are empty.
  • Prevented Internal Users from Creating Invalid Keys.
  • Fixed Key Alias Re-validation on Update blocking legacy aliases.
  • Registered DynamoAI guardrail initializer and enum entry.
  • Skipped #transform=inline for base64 data URLs in Fireworks provider.
  • Avoided no running event loop during synchronous initialization for Langsmith.
  • Fixed Gemini provider to support images in tool_results for /v1/messages routing.
  • Fixed CSV export being empty on the Global Usage page.
  • Corrected supported_regions for Vertex AI DeepSeek models in model-prices.
  • Restored gpt-4-0314 pricing information.
  • Fixed Redis cluster caching.
  • Converted max_budget to float when set via environment variable in proxy.
  • Mapped Anthropic 'refusal' finish reason to 'content_filter'.
  • Fixed Vertex AI streaming finish_reason for gemini-3.1-flash-lite-preview to be 'stop' instead of 'tool_calls'.

Affected Symbols