Google Makes Gemini 3.5 Transcribe and Omni Flash 1.1 Generally Available

Google has moved three important Gemini API capabilities into general availability: Gemini 3.5 Transcribe, Gemini 3.5 Transcribe Live, and Gemini Omni Flash 1.1. The changes matter less as isolated model-name updates and more as a production signal: speech transcription and multimodal generation workflows that had been preview-oriented now have GA endpoints designed for broader deployment.

For speech, the new generally available models are gemini-3.5-transcribe and gemini-3.5-transcribe-live. Google's documentation says the transcription stack supports more than 85 languages and locales, automatic language detection, code-switching, speaker diarization and word-level timestamps. Developers can also supply custom vocabulary lists of up to 1,000 terms, which is useful for domain-specific names, technical language and organization-specific terminology.

The live model is aimed at streaming use cases. Google documents WebSocket-based streaming with interim and final transcription events, plus multiple voice-activity-detection strategies. Its Smart transcription mode is intended to help balance responsiveness and transcript stability instead of forcing every application to implement that behavior independently. Those capabilities make the release relevant to meeting assistants, call-center workflows, live captions, voice interfaces and real-time agent systems.

Google also made Gemini Omni Flash 1.1 generally available under the model ID gemini-omni-1.1-flash. The Omni line is designed around multimodal generation and transformation. Current documentation highlights video extension, interpolation using first and last frames, and a resolution control supporting 360p, 720p, 1080p and 4K outputs. Google notes that 1080p and 4K are upscaled outputs, an important implementation detail for teams comparing generation quality or estimating downstream media-processing requirements.

The release notes also create a migration deadline. Google says the earlier gemini-omni-flash-preview model is deprecated and scheduled for removal on September 30, 2026. Teams using the preview should therefore test the GA endpoint, compare output behavior and update model routing before that date rather than treating the preview endpoint as a durable production dependency.

The larger pattern is that Google is productizing two different parts of the multimodal stack at once. Transcribe and Transcribe Live focus on converting speech into structured, application-ready text, while Omni Flash 1.1 expands generation and transformation workflows across media. Together they make it easier to build pipelines that listen, understand, generate and transform content without stitching together as many separate model families.

These releases are generally available, not merely announced or in preview. That distinction is important for production planning. GA does not remove the need for application-level testing, latency measurement, cost controls, privacy review or fallback handling, but it changes the maturity signal developers should use when deciding which Gemini endpoints belong in production architectures.

For teams already using the preview models, the practical next steps are straightforward: inventory model IDs, test transcript accuracy and streaming behavior on representative audio, validate custom vocabulary and diarization needs, benchmark Omni output at the resolutions actually required, and complete migration from the deprecated Omni preview before September 30, 2026.