LiteLLM 1.108.3 is a signed LTS release that repairs Bedrock requests when api_base is an empty string, allows MCP tool results larger than 1 MiB through the SSE path and prevents duplicate model-level post-call guardrail moderation. It follows the 1.108.2 credential and transcript-redaction fixes on the same retained release line.
Why it matters: These fixes sit on three production boundaries: where a Bedrock request is sent, whether a large tool result can complete and whether a moderation policy executes once or twice. Each can change reliability, cost or enforcement without changing application…
NVIDIA and Microsoft released enterprise guidance and evaluation resources for open-weight models. The framework separates model risk from harness and runtime risk, maps five risk categories to 14 evaluations, and proposes a five-step process from scope and acceptance criteria through mitigation, retesting and continuous reassessment.
Why it matters: Self-hosting transfers safety, data protection and monitoring responsibilities from an API provider to the deployer. The guide provides a concrete starting inventory while explicitly warning that a model-only test cannot establish the safety of an agentic…
LiteLLM 1.106.2 is a signed stable patch that fixes Bedrock Mantle count_tokens 404 responses and several batch cost, pricing, translation and spend-accounting defects, including a P0 batch-cost issue. It extends the canonical 1.106 release event with billing-correctness fixes.
Why it matters: Incorrect token and batch accounting can distort spend controls and customer chargeback even when inference succeeds. The patch is operationally material for gateways that use LiteLLM as their cost ledger.
AWS made Claude Opus 5.5 and Claude Sonnet 5.5 available on Amazon Bedrock in AWS GovCloud (US). The bedrock-runtime endpoint is available in US-East and US-West with Guardrails and invocation logging; the Anthropic-compatible bedrock-mantle endpoint is available in US-West with server-side tools, background inference and Projects.
Why it matters: Regulated teams now have a GovCloud path for newer Claude models, but endpoint choice changes the control surface. Logging and Guardrails live on bedrock-runtime, while Mantle offers Anthropic-native features, so compliance evidence cannot be inferred from…
LiteLLM promoted 1.105 to signed release candidate 1. The candidate retains the earlier agent-permission, delegated-identity, credential-isolation and tracing work while adding a Microsoft 365 Graph MCP catalog entry, non-canonical activity-date rejection, forced exit after database setup failure, fail-closed Straiker v3 handling when a verdict is missing, and storage, guardrail and migration fixes.
Why it matters: The release candidate is a more testable checkpoint for the gateway's agent identity and policy work, and several additions move error handling from ambiguous continuation toward explicit failure. It remains a preview with a broad change surface, so…
LiteLLM 1.101.0 promoted the release-candidate line to stable with per-user semantic-cache isolation, credential redaction, MCP identity checks, admission control, budget alerts and routing controls. Stable also adds OpenAI workload-identity federation, a Responses input-token counting endpoint, modality and context-window routing, and several provider and guardrail fixes.
Why it matters: Teams waiting for a stable checkpoint can now test the 1.101 controls without adopting a development tag. Stable labeling reduces lifecycle uncertainty but does not prove tenant isolation, cost accounting or provider parity in a specific deployment.
LiteLLM 1.100.1 patches the stable 1.100 line so retry breadcrumbs stop retaining every earlier request and removes spend-attribution backports from that branch. The signed release retains 1.100's governed routing, agent-registry search, access-group budgets, MCP toolset enforcement, guardrail logging and job-ownership controls.
Why it matters: The patch narrows unnecessary request-history retention in retry paths and changes which spend-attribution fixes operators can assume are present on the stable line.