Gemini Live — primary 2.5 and BETA 3.8

Gemini Live 2.5 Flash Native Audio — Primary is the default voice model for new bots. BeAI no longer labels it deprecated. There is no scheduled automatic migration from 2.5 to 3.8.

Gemini 3.8 Live — BETA and Gemini 3.8 Live Extended Thinking — BETA remain available by explicit selection. BETA describes their validation status within BeAI, independently of Google's published status. Bots already using these models retain their selection.

Choose a model

In Bots → Edit → Voice and AI model, select the model and a compatible voice, then save. Changes apply to new calls. Bot templates and PicApp solutions share the catalogue; an existing explicit configuration takes precedence over the default.

Bot management permissions, organisation scope and hidden or locked PicApp fields still apply. Each model requires an authorised positive organisation price. Primary and BETA statuses change neither prices nor these checks. A model without authorised pricing remains unavailable; BeAI does not silently select another model.

The legacy Gemini 3.1 Flash Live Preview identifier still maps to 3.8 Live for compatibility. This does not apply to Gemini Live 2.5. Conversation history retains the model actually used.

Settings and limits

Both 3.8 variants support the bot's authorised tools. Extended Thinking uses LOW reasoning and may continue processing after an audio sentence. Longer reasoning can increase consumption costs.

Audio proactivity is always enabled on 3.8, so its switch is not offered. It remains configurable on 2.5. Other advanced speech detection settings remain available. An incompatible voice is rejected on save. Provider session failures (credentials, model access or quotas) appear as AI startup failures in conversation history.

Public API

POST /api/v1/bots without ai_model also defaults to gemini-live-2.5-flash-native-audio, subject to authorised pricing. Change an existing bot with PATCH /api/v1/bots/{id}/voice:

{"ai_model":"gemini-live-2.5-flash-native-audio","voice":"Puck"}

The BETA identifiers are gemini-3.8-live and gemini-3.8-live-extended-thinking. model_availability and GET /api/v1/organisations/{id}/model-pricing expose is_default, beta and deprecated alongside pricing and availability. These values are true, false, false for 2.5 and false, true, false for both 3.8 variants. GUI and API permissions and restrictions are identical.

See Voice model pricing and Gemini consumption.