Google Cuts Gemini Omni 1.2 Flash API Price 30% and Extends Clips to 15 Seconds

Google DeepMind released Gemini Omni 1.2 Flash on November 5, 2026, cutting the 720p API price by 30 percent to $0.07 per second while extending the maximum native clip length from 10 seconds to 15 seconds — the two most-cited practical barriers to adopting Omni over one-shot video systems like Veo 3.1. The update is generally available across the Gemini API, Google Cloud Vertex AI, and Google AI Studio, with the model identifier gemini-omni-1.2-flash.

The simultaneous price and capability upgrade strengthens Omni's position not only against external competitors but also against Google's own Veo 3.1 — which produces higher raw fidelity but scores roughly 150 points lower on the Arena.ai human-preference video leaderboard.

How Gemini Omni 1.2 Flash Differs From One-Shot Video Models

Gemini Omni is architecturally distinct from systems like Veo or Runway Gen-4 that generate a single clip from a text prompt and return a finished file. Omni processes text, images, audio, and video in one unified forward pass rather than routing them through specialized encoder and decoder sub-systems. The practical effect is that every generation is condition-aware across all input modalities simultaneously: a prompt describing a character, a reference photograph, and an instruction to change the lighting are all evaluated in the same model execution rather than sequentially.

The more significant architectural distinction in version 1.2 concerns the editing pass. In Gemini Omni 1.1 Flash, conversational editing — asking the model to revise a clip you just generated — was constrained to analyzing only the final second of the existing video. Omni 1.2 Flash retains and processes the full clip across editing turns, so when a user asks it to adjust the camera angle in the middle of a 12-second sequence, the model understands the entire temporal context rather than extrapolating from an endpoint, according to Google.

The extended context window for editing is what makes the increase from 10-second to 15-second clips practically useful: without full-clip context retention during edits, longer clips would accumulate inconsistencies across revision turns. The two changes are engineered together.

Production Economics: $0.07 Per Second Changes the Calculation

Gemini Omni 1.1 Flash charged $0.10 per second for 720p output. Version 1.2 drops that rate to $0.07 per second — making a 15-second clip $1.05 at 720p, the same amount that bought 10 seconds under the previous pricing. That per-clip price point puts programmatic video generation within range of marketing workflows that run dozens of creative variants simultaneously.

Google explicitly listed Google Ads as a deployment integration for Omni 1.2, alongside Google Vids, Google Flow, and the Gemini app. The Ads integration is commercially significant: at $1.05 per clip, A/B testing five or ten video creative variants becomes a line item in a mid-market advertising budget rather than an experimental research cost. If Omni 1.2 becomes a standard input for programmatic ad systems, it represents a structural shift in how AI-generated video enters production — not through editorial workflows but through ad-platform automation.

Consumer access to the full 15-second clip length is restricted to Google AI Ultra subscribers in the Gemini app. Developers accessing the model directly through the Gemini API or Vertex AI have unrestricted access to the full 3-to-15-second range in one-second increments.

Where Gemini Omni 1.2 Stands on the Arena.ai Leaderboard

The Arena.ai video leaderboard, operated by UC Berkeley's Chatbot Arena team, scores AI video models using an Elo-style Bradley-Terry rating computed from blind pairwise human preference votes. Because it captures aesthetic preference rather than automated pixel-level metrics, Arena scores often diverge from automated benchmark rankings — which is why Veo 3.1, optimized for technical fidelity, scores roughly 1,364 on the platform while Gemini Omni Flash, optimized for iterative usability, led the text-to-video category with a score of 1,516 for version 1.1 as of late September 2026.

Google claims Gemini Omni 1.2 Flash now holds the top text-to-video ranking with a score of 1,541. As of publication, that specific score has not been confirmed from an independently available leaderboard snapshot — the most recent confirmed data reflects 1.1 Flash at 1,516, and leaderboard scores update continuously as new votes accumulate following a model launch. The claim is consistent with the typical score progression when a significantly improved version of an already-leading model enters the arena.

The leaderboard position applies specifically to the text-to-video category. In image-to-video evaluation, the Omni family ranks lower — MiniMax H3 and several Seedance variants hold stronger positions in that sub-category — a gap that 1.2's improved reference image adherence is specifically designed to narrow.

Competitive Position: Runway, Kling, and a Narrowing Gap

The competitive field shifted substantially in April 2026 when OpenAI shut down Sora, removing the most prominent alternative text-to-video platform. The current market now centers on Runway Gen-4.5, Kling 3.0, ByteDance's Seedance 2.0, and MiniMax Hailuo 2.3 as the primary independent alternatives to Google's video generation stack.

None of those systems natively combine multimodal input processing with multi-turn conversational editing in the way Gemini Omni does. Runway Gen-4.5 and Kling 3.0 offer editing tools, but they operate on completed clips rather than through a conversational model that retains full generation context. The conversational architecture is Omni's clearest technical differentiation from the competitive set — and it is the reason Arena.ai human voters, who are evaluating usability across interaction cycles rather than single-shot outputs, have consistently ranked Omni above models with comparable raw video quality.

The 30% price cut also repositions Omni against Veo 3.1 Fast, Google's lower-latency video generation mode, which carries a $0.10-per-second 720p price on Google Cloud Vertex AI. Veo 3.1 preview model IDs were deprecated from the Gemini API on October 22, 2026, with Gemini Omni 1.1 Flash designated as the replacement on that platform; Veo 3.1 Fast remains available through Vertex AI for existing integrations. For use cases that do not require Veo's one-shot fidelity — iterative ad creative, rapid pre-visualization, marketing content that will be refined through multiple editing turns — Omni 1.2 is now cheaper than Veo Fast while returning higher Arena preference scores.

Limitations and Claims Requiring Independent Confirmation

Several aspects of the 1.2 release require additional verification. The claimed Arena.ai score of 1,541 for the new version has not been independently confirmed from a live leaderboard snapshot at time of publication. Resolution tiers beyond 720p — including the 1080p and 4K output options supported in version 1.1 — have not had updated pricing confirmed for 1.2 from primary documentation reviewed.

Content moderation strictness remains a documented practical friction: reviewers of prior Omni versions noted frequent generation refusals, and there is no indication from the 1.2 announcement that the guardrail calibration has changed. SynthID watermarking is embedded invisibly in all Omni outputs, which limits the model's utility in contexts requiring unmarked synthetic video for professional post-production. Regional restrictions apply to some avatar generation features.

Omni 1.2 Flash reduces the total extension ceiling to 30 seconds, down from the 40-second limit supported in version 1.1 Flash, with each editing turn extending the existing clip by 3 to 10 seconds per call. Developers building sequences longer than the native 15-second maximum should account for this revised limit when planning iterative generation workflows.

What to Watch Next

The most consequential near-term development is independent evaluation of Omni 1.2 Flash's reference consistency improvements. The model's ability to hold character appearance, brand assets, and visual continuity across editing turns determines how far it can penetrate professional production workflows that currently rely on manually operated tools — a market where current Arena preference scores alone are insufficient evidence.

The Google Ads integration rollout schedule also warrants attention. If the model becomes a standard creative input for programmatic advertising at scale, it will generate pricing and performance data that will either validate or complicate the $0.07-per-second economics under real production load conditions. Pricing disclosures for 1080p and 4K output tiers — absent from public documentation as of launch — will determine whether the cost advantage over competitors extends beyond the 720p tier where the comparison is currently clearest.