Live API
A real-time voice and vision API provided by Google DeepMind, based on stateful WebSocket, that supports multimodal streaming sessions, interruption and tool calling.
Timeline (last 90 days)
The impact of the breaking change in v3.37.0 is concentrated on the Decisions type of the Live API.
For projects using the Live API and the Decisions type, it is recommended to globally search for string operations on question.instructions before upgrading, check exhaustive judgments on Decisions input message roles, and add test cases for structured instructions.
In v3.37.0, the instructions field of the question in the Live response model can now be either a string or a list composed of structured text, images, or JSON fragments.
The Live API is an interface in the Gemini API for real-time bidirectional interaction, based on WebSocket or WebRTC long connections, continuously receiving audio/video streams and immediately returning voice responses.
Google's official announcement states that two Live API preview models and three TTS preview models will be officially shut down on November 17, 2026.
During migration, the Live API's session initialization parameters (such as audio sample rate, voice activity detection, and context window configuration fields) and the TTS request body (such as voice selection, output format, and rate parameters) need to be adjusted.
The Live API's billing model based on session duration is usually measured in seconds or minutes that the connection is maintained, meaning that even if the user is silent, charges may be incurred as long as the connection remains open.
Key features of the Live API include support for 90 languages, barge-in, tool-calling capability and audio transcription.
The Live API is a low-latency real-time voice and vision API with a stateful WebSocket design: after a developer creates a session, they can continuously send text, audio and video frames over the WebSocket, and the model streams audio chunks back in real time along with transcript text.
OpenAI's Realtime API, ElevenLabs' low-latency mode, and Google's Live API all emphasize low-latency voice capabilities.
1 more timeline events
All Facts (11)
The impact of the breaking change in v3.37.0 is concentrated on the Decisions type of the Live API.
50%UnverifiedFor projects using the Live API and the Decisions type, it is recommended to globally search for string operations on question.instructions before upgrading, check exhaustive judgments on Decisions input message roles, and add test cases for structured instructions.
50%UnverifiedIn v3.37.0, the instructions field of the question in the Live response model can now be either a string or a list composed of structured text, images, or JSON fragments.
50%UnverifiedThe Live API is an interface in the Gemini API for real-time bidirectional interaction, based on WebSocket or WebRTC long connections, continuously receiving audio/video streams and immediately returning voice responses.
50%UnverifiedDuring migration, the Live API's session initialization parameters (such as audio sample rate, voice activity detection, and context window configuration fields) and the TTS request body (such as voice selection, output format, and rate parameters) need to be adjusted.
50%UnverifiedThe Live API's billing model based on session duration is usually measured in seconds or minutes that the connection is maintained, meaning that even if the user is silent, charges may be incurred as long as the connection remains open.
50%UnverifiedThe Live API is a low-latency real-time voice and vision API with a stateful WebSocket design: after a developer creates a session, they can continuously send text, audio and video frames over the WebSocket, and the model streams audio chunks back in real time along with transcript text.
50%UnverifiedKey features of the Live API include support for 90 languages, barge-in, tool-calling capability and audio transcription.
50%UnverifiedOpenAI's Realtime API, ElevenLabs' low-latency mode, and Google's Live API all emphasize low-latency voice capabilities.
50%UnverifiedDevelopers can access these two models through the Live API.
50%UnverifiedGoogle's official announcement states that two Live API preview models and three TTS preview models will be officially shut down on November 17, 2026.
50%