Gemini 2.5 Flash Exp Native Audio Thinking Dialog
Google released Gemini 2.5 Flash Exp Native Audio Thinking Dialog for the Live API on May 20, 2025. This brief details verified modalities and evaluation steps.
Verified Release Facts
The official changelog records that Google released gemini-2.5-flash-preview-native-audio-dialog and gemini-2.5-flash-exp-native-audio-thinking-dialog on May 20, 2025. These are identified as new Gemini models specifically designed for the Live API. The documentation states they possess native audio output capabilities. The input modalities listed for these models include text and audio, while the output modalities include text and audio. The provider is identified as Google, and the model family is Gemini.
- Release date: May 20, 2025.
- Models: gemini-2.5-flash-preview-native-audio-dialog and gemini-2.5-flash-exp-native-audio-thinking-dialog.
- Input modalities: Text and audio.
- Output modalities: Text and audio.
- Open weights status: Not confirmed (openWeights=false).
Limitations and Unknowns
The source material does not specify pricing, context window size, or maximum output tokens for these models. It does not confirm whether local installation is possible or if API access is mandatory. The fact that open weights are not confirmed does not establish that local deployment is impossible. Additionally, the source does not describe the internal architecture or how the 'thinking' feature operates. There is no information regarding performance rankings or comparisons against previous iterations. A later changelog entry notes that gemini-2.5-flash-exp-native-audio-thinking-dialog has been shut down and suggests using gemini-2.5-flash-native-audio-preview-09-2025 instead, indicating a change in availability over time.
Practical Evaluation Procedure
Engineers can evaluate these models by conducting controlled tests within the Live API environment. First, test the models by sending specific audio prompts to verify that native audio output is generated correctly. Measure the latency between the end of the user's audio input and the start of the model's audio response. Compare the audio quality and responsiveness against baseline expectations defined by your application requirements. Validate that the text input modality functions as expected when mixed with audio streams. Do not assume operational benefits such as cost savings or speed improvements without direct measurement. Record all results to determine if the model meets specific technical criteria for your use case.
Primary sources
Frequently asked questions
What are the confirmed input and output modalities?
The source confirms that both models accept text and audio as inputs. They produce text and audio as outputs. No other modalities are listed in the provided facts.
Is the model available for local installation?
The source does not state whether local installation is possible. It only records that open weights are not confirmed. Therefore, no conclusion can be drawn about local deployment requirements from this data alone.