Gemini Omni Flash API is live: public preview, $0.10/sec, and what builders should know
As of June 30, 2026, gemini-omni-flash-preview is in public preview on the Gemini API, Google AI Studio and Gemini Enterprise Agent Platform — with conversational editing, a 10-second cap and $0.10 per second pricing.
The wait is over
At Google I/O 2026 Google said a developer API for Gemini Omni Flash was coming “in the coming weeks.” On June 30, 2026 that surface actually shipped: model ID gemini-omni-flash-preview is available in Google AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform.
Consumer surfaces (Gemini app, Google Flow, YouTube Shorts / Create) already had Flash. This post is about the developer moment — pricing, limits, and what still is not ready.
Official launch write-up: Start building with Nano Banana 2 Lite and Gemini Omni Flash.
What you can call today
| Item | Value |
|---|---|
| Model ID | gemini-omni-flash-preview |
| Status | Public preview |
| Video output price | $0.10 / second (same as Veo 3.1 Fast) |
| Max clip length | 10 seconds (longer durations coming) |
| API resolution | 720p |
| Inputs | Text, image, video |
| Outputs | Video (+ text responses) |
| Editing | Conversational, multi-turn |
| Provenance | SynthID watermark + C2PA Content Credentials |
Omni Flash is positioned as the cost-efficient video workhorse in the Omni family: multimodal references, Gemini-grounded world knowledge, and chat-style edits without regenerating the whole clip from scratch.
Where it runs
Consumers
- Gemini app (Google AI Plus / Pro / Ultra)
- Google Flow
- YouTube Shorts and YouTube Create (free for 18+)
Developers
- Google AI Studio playground
- Gemini API
- Gemini Enterprise Agent Platform
Pricing in practice
Google priced Omni Flash video output at $0.10 per second. A full 10-second clip is about $1.00 of output before input tokens. Preview pricing does not currently advertise a Batch discount, so there is no “cheaper if you can wait” lever on this model yet.
Consumer quotas still follow Google AI subscription compute limits. Do not mix those with API billing when you model agency or product unit economics.
Limits you should design around
Google is explicit about preview constraints:
- 10-second generations — hard product cap for now; longer clips are on the roadmap.
- No API audio references — upload audio is not supported for this model on the Gemini API yet (output can still carry native synced audio).
- Short video references are brittle — the schema may accept video references up to 3 seconds, but Google notes they are not correctly processed at this time.
- Scene changes and pans — character consistency can still slip across cuts; plan for extra edit turns or tighter references.
Treat these as product constraints, not temporary support tickets.
Omni Flash API vs Veo 3.1
| Factor | Omni Flash (preview) | Veo 3.1 |
|---|---|---|
| Architecture | Multimodal Omni family | Specialised video model |
| Editing | Native conversational edits | Re-prompt / regenerate |
| Price signal | $0.10 / sec output | Veo Fast matches $0.10 / sec; higher tiers cost more |
| Clip length | Up to 10s | Often ~8s in production docs |
| Maturity | Public preview | Generally available |
If you need a stable production path today, Veo 3.1 still matters. If you need in-chat editing and Omni-style multimodal steering, Flash preview is the new surface. The clean pattern is still one VideoProvider interface with two implementations — see our earlier API developer guide for the migration shape (that guide was written before the preview dropped; this article is the status update).
Pairing with image models
Google’s June 30 post also launched Nano Banana 2 Lite for fast, cheap image drafting. The intended loop is simple: generate or edit stills quickly, then pass a chosen frame into Omni Flash as an image-to-video reference. Demo apps on AI Studio show that pattern end to end.
Bottom line
Gemini Omni Flash is no longer “API soon.” As of June 30, 2026 you can build against gemini-omni-flash-preview at $0.10/sec, with a 10-second cap and clear preview limits on audio references and short video refs. Ship conversational-edit flows behind a provider interface, keep Veo for workloads that need GA stability, and watch Google’s docs for longer durations and audio-reference support.