LiteLLM's router can mirror live traffic to several shadow models at once.

v1.103.0-dev.2 adds streaming shadow traffic and fans silent_model out to multiple targets, so you can evaluate two or three candidates against real production prompts without serving their output. The same release bounds the content filter to a per-chunk window instead of rescanning, recounts tokens when a streamed Responses call completes without usage, and tolerates Anthropic message_delta events with no usage field.