Gemini 2.5 Flash Native Audio Preview Release
Google released Gemini 2.5 Flash Native Audio Preview 12-2025 on December 12, 2025. This native audio model supports text and audio inputs via the Live API.
Official Release Details
Google announced the release of gemini-2.5-flash-native-audio-preview-12-2025 on December 12, 2025. This specific date marks the historical event of the model's availability in the official changelog. The article publication date is distinct from this announcement date.
The model is identified as a new native audio model intended for use with the Live API. According to the official documentation, this update improves the model's ability to handle complex workflows. The source does not specify which particular workflows are improved or how they differ from previous iterations.
Technical Specifications and Limitations
The model accepts both text and audio as input modalities and generates both text and audio as output modalities. The documented context window is 131,072 tokens, while the maximum output length is set at 8,192 tokens.
Open weights are not confirmed for this release. This status indicates that open weights are not verified; it does not establish that local installation is impossible, nor does it confirm that API access is mandatory. Pricing information is currently null in the source data, meaning no cost structure is provided. No architectural details, such as internal mechanisms or deployment requirements, are specified in the available facts.
- Input Modalities: Text, Audio
- Output Modalities: Text, Audio
- Context Window: 131,072 tokens
- Max Output Tokens: 8,192
- Open Weights: Not Confirmed
Evaluation Procedure for Engineers
Engineers can evaluate the model by designing specific tests within the Live API environment. A practical procedure involves creating a baseline test case involving a complex workflow scenario defined by the user. Measure the model's response latency and accuracy against this baseline. Compare the results with previous models if applicable data exists, but note that the source does not provide comparative benchmarks.
Validate the output quality by checking if the audio responses align with the expected semantic content of the text prompts. Since the source states the model handles complex workflows, engineers should test multi-step interactions to observe stability. However, these steps are proposed tests; they do not constitute conclusions about the model's inherent capabilities beyond what is officially stated.
Primary sources
Frequently asked questions
What is the primary function of this new model?
The model is a native audio model for the Live API. Its primary documented function is to improve the ability to handle complex workflows while accepting text and audio inputs and producing text and audio outputs.
Is local installation possible for this model?
The source records that open weights are not confirmed. This does not establish that local installation is impossible, nor does it confirm that API access is mandatory. No hosting arrangement or deployment architecture is specified in the provided facts.