Gemini Omni 1.1 Flash and 3.5 Transcribe Push Multimodal AI Toward Production APIs

Two multimodal capabilities moved from experimentation toward production

Google’s Gemini API release notes show two important general-availability changes on consecutive days in late August 2026. On August 27, Google released gemini-omni-1.1-flash as the GA version of its conversational video generation and editing model. The release adds video extension, first-and-last-frame interpolation and explicit resolution controls for 360p, 720p, 1080p and 4K output, with the higher resolutions produced through upscaling. Google also says the older gemini-omni-flash-preview endpoint is scheduled for deprecation on September 30, 2026.

The production implication is not merely better-looking video. Developers can now treat iterative editing, scene continuation and frame-constrained transitions as API-level operations rather than stitching together separate experimental tools. Google’s developer announcement describes Omni 1.1 Flash as production-ready, while Google Flow is already exposing the same control pattern for creative users.

Speech transcription is becoming a dedicated model family

One day earlier, on August 26, Google made Gemini 3.5 Transcribe and Gemini 3.5 Transcribe Live generally available. The non-streaming model supports utterance-level language detection across more than 85 languages, speaker diarization, word-level timestamps and vocabulary biasing for up to 1,000 terms. The Live variant adds low-latency bidirectional streaming over WebSockets with interim and finalized transcription events plus multiple voice-activity-detection strategies.

That separation matters for product teams. Instead of using a general multimodal model for every audio workload, developers now have a dedicated transcription path with production-oriented features such as timestamps, speaker handling and custom terminology, while the live endpoint targets call, meeting and real-time assistant experiences.

What developers should watch

The clearest migration signal is the preview retirement date for Omni Flash. Teams using the preview endpoint should test the GA model before September 30 and verify output-resolution, latency and workflow differences. Speech teams should compare the batch and Live transcription paths against their latency, diarization and vocabulary needs rather than assuming one endpoint fits every product.

The broader trend is that multimodal AI is being decomposed into production services with clearer task boundaries: controllable video generation/editing on one side and specialized streaming or batch transcription on the other. That makes multimodal systems easier to integrate, but it also increases the importance of model-version tracking, deprecation monitoring and workload-specific evaluation.