Member Junction
    Preparing search index...

    Interface RealtimePricingSettings

    Prices for realtime usage that the model's cost rows (MJ: AI Model Costs) do not price on their own: avatar video, and audio tokens, which providers bill at a different rate from the text tokens a cost row prices. The cost rows stay the authority for every other token price. A price belongs to the host that charges it, so set this on the model-vendor row: a price on the model or the model type would reach every vendor that serves the model.

    interface RealtimePricingSettings {
        AudioInput?: RealtimeUnitPrice;
        AudioOutput?: RealtimeUnitPrice;
        AvatarVideoOutput?: RealtimeUnitPrice;
    }
    Index

    Properties

    AudioInput?: RealtimeUnitPrice

    The price of audio input tokens, charged on a line of its own instead of at the cost row's input rate. Realtime providers bill audio tokens at a different rate from text tokens (GPT Realtime: several times the text rate), and a cost row holds one input rate. 'Per 1M Tokens' only; any other unit is not applied. Cached audio input tokens are priced at the cost row's cache-read rate, not here. Without this price, audio input tokens are priced at the cost row's input rate and the run's default cost line is marked as blended.

    AudioOutput?: RealtimeUnitPrice

    The price of audio output tokens, charged on a line of its own instead of at the cost row's output rate. 'Per 1M Tokens' only; any other unit is not applied. Without this price, audio output tokens are priced at the cost row's output rate and the run's default cost line is marked as blended.

    AvatarVideoOutput?: RealtimeUnitPrice

    The price of avatar video output, charged on top of the session's token cost. 'Per Minute' prices the seconds of avatar video the model generates; 'Per 1M Tokens' prices the output tokens the provider counts as video. Set the price and its unit together: the cascade merges per key, so a layer that sets only one of them keeps the other from the layer below.

    Gemini 3.8 Live on Vertex AI is seeded at Google's $1.00 per 1M avatar video output tokens. Google bills avatar video only while the avatar speaks, so a token price follows the bill; a per-minute price would also count the idle avatar between answers.