v1.83.10-nightly
Breaking Changes📦 litellmView on GitHub →
⚠ 1 breaking✨ 8 features🐛 17 fixes🔧 23 symbols
Summary
This release introduces mandatory Python 3.10+ support and enhances security by implementing Docker image signing verification via Cosign. New features include support for grok-4.20-0309-reasoning and improved budget management capabilities.
⚠️ Breaking Changes
- Python 3.9 is no longer supported. The minimum required Python version is now 3.10 (requires-python changed from >=3.9, <3.14 to >=3.10, <3.14).
Migration Steps
- Ensure your Python environment is running version 3.10 or higher, as Python 3.9 is no longer supported.
✨ New Features
- Added support for grok-4.20-0309-reasoning model.
- Implemented per-member model scope + team default_team_member_models for teams.
- Added capability for multiple concurrent budget windows per API key and team.
- Added capability to override default GitHub Copilot authentication endpoint.
- Added support for Claude Code BYOK in the UI Settings.
- Implemented support for Mistral-style pages param via analyze query string for ocr/azure-di.
- Added tpl support to extraContainers and extraInitContainers in helm charts.
- Implemented configurable multi-threshold budget alerts for virtual keys.
🐛 Bug Fixes
- Fixed stale managed object cleanup in batch-limit to prevent 300K row UPDATE.
- Fixed /v1/messages/count_tokens endpoint to return the actual status code.
- Restricted x-pass- header forwarding for credential and protocol headers.
- Tightened api_key value check in credential validation.
- Added missing Mistral OCR params to allowlist for OCR.
- Fixed return value for routing_strategy_args when strategy is not latency-based (returns None).
- Fixed enforcement of organization boundaries in proxy admin operations.
- Fixed proxy to read guardrail config from admin metadata and fixed tag routing consistency.
- Added URL validation for user-supplied URLs in the proxy.
- Prevented negative streaming costs for start-only cache usage in bedrock.
- Fixed proxy to prioritize reasoning health-check max token precedence.
- Persisted default router end budget.
- Fixed extra_headers not persisting on MCP server edit in UI.
- Fixed UI to strip empty premium fields from key update payload.
- Resolved prometheus_helpers file/package shadow breaking /global/spend/logs.
- Enforced project-level model-specific rate limits in parallel_reqs.
- Fixed MCP to restore PKCE-triggering 401 when no stored per-user token exists.
Affected Symbols
batch-limit/v1/messages/count_tokensx-pass- headerapi_keyMistral OCR paramsrouting_strategy_argsutils/pkce.tsproxy admin operationsguardrail configtag routinguser-supplied URLsgrok-4.20-0309-reasoningGitHub Copilot authentication endpointbedrock streaming costsreasoning health-check max tokenrouter end budgetextra_headersMCP server editpremium fieldsprometheus_helpers/global/spend/logsparallel_reqsPKCE-triggering 401