Thanks — I’m following up about the issues with gemini-3.1-flash-live-preview. • Status: A long-term fix is actively being investigated. • Known limitations right now: ◦

send_client_content

only works for initial history seeding; later calls return a 1007 error. ◦

generate_reply()

,

update_instructions()

, and

update_chat_ctx()

are ignored (session state updates internally but those calls are not applied). ◦ Agent handoffs with

session.update_agent()

don’t work mid-session. ◦ Affective dialog and proactive audio are not supported. ◦ Asynchronous function calling is not supported — the model waits for tool responses before continuing. ◦

thinkingConfig

now uses

thinkingLevel

(

"minimal"

,

"low"

,

"medium"

,

"high"

) instead of

thinkingBudget

. • What works: Basic voice conversations, tool calling, and audio I/O operate normally. No ETA is available yet. If you share how you’re using the model (sample workflow, which APIs you call, and whether you rely on mid-session instruction updates or async tool calls), I can suggest workarounds or migration steps tailored to your use case. Sources: Gemini Live API plugin | LiveKit Documentation, with interactive elements