LiteLLM v1.104.0-dev.1
LiteLLM v1.104.0-dev.1 introduces authentication security enhancements (breached password checks, self-service resets), proxy improvements including 429 rejections for saturated deployments, Amazon Transcribe pass-through, and extensive UI/Rust refactoring.
Why this signal matters
LiteLLM v1.104.0-dev.1 brings several operational, security, and performance updates: - Authentication & Security: Adds breached password detection to password policy, self-service password changes, forced resets for breached or admin-configured passwords, and resolves plaintext password exposure. - Router & Proxy: Enforces 429 rate limit errors when deployment parallel request capacity is exhausted; defers fastapi and tiktoken imports on library load; optimizes /model/info via single orjson serialization; shifts prompt injection heuristics off the async event loop; updates proxy configuration precedence so config files override database settings; and evicts JWT key mapping cache during entity deletions. - Integrations & MCP: Adds Amazon Transcribe pass-through with completion-time job pricing, support for streaming Vertex AI GCS batch outputs, distinct HashiCorp Vault namespaces for login and secrets, and visibility into live MCP gateway sessions per client/user. - Pricing & Budgeting: Syncs OpenRouter model pricing catalogs, implements fallback billing for cache-read tokens using input rates when unspecified, fixes Bedrock Titan batch token billing, introduces temporary budget increments for team members, and sets per-key default budgets for dynamic customers.
Actionable summary
LiteLLM release v1.104.0-dev.1 adds breached password detection, self-service password reset, and enforced reset flows. The proxy router now returns 429 status codes when max parallel request slots are saturated, optimizes /model/info serialization with orjson, offloads prompt injection heuristics from the event loop, and defers fastapi/tiktoken imports. It also adds MCP live gateway session tracking, HashiCorp…
- Agent usefulness
- 82/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Verify container images using the provided Cosign public key hash ( 0112e53046018d726492c814b3644b7d376029d0 ) or tag before deployment.
- Review authentication configurations to prepare users for forced password resets if using breached or admin-set credentials.
- Verify client-side retry handling for HTTP 429 responses resulting from saturated parallel request limits on deployments.
- Check deployment configurations if relying on database overrides, as proxy configuration files now take precedence over database settings.