Gemini Omni 1.2 Flash Drops to $0.07 a Second at 720p and Stretches Clips to 15 Seconds

Google cut the price of Gemini Omni video generation to $0.07 per second at 720p on Nov. 5, down from about $0.10, and raised the longest single clip from 10 seconds to 15. A 15-second 720p clip now costs about $1.04, according to Google's pricing page.

Anish Nangia, a product manager at Google DeepMind, announced Gemini Omni 1.2 Flash in a blog post. The model ID is gemini-omni-1.2-flash, and it is generally available on the Gemini API and on Google Cloud.

What Gemini Omni 1.2 Flash costs

Google's pricing page bills the model on video output tokens at $12.00 per million. Gemini Omni 1.1 Flash was $17.50 per million. The token rate per second of video did not change, so the entire cut comes from the per-token price. That works out to a 31% reduction, which is this publication's arithmetic and matches the 30% Google claims.

Resolution	Tokens per second	Price per second
360p	1,931	$0.023
720p	5,792	$0.07
1080p, upsampled	8,688	$0.104
4K, upsampled	17,376	$0.209

The 720p figure is $0.0695 before rounding. Google's blog says drafts can be generated at 360p or 720p and then upscaled to 1080p or 4K, but does not price that step. The pricing page lists 1080p and 4K only "with upsampling."

1. 360p0.02 USD
2. 720p0.07 USD
3. 1080p (upsampled)0.1 USD
4. 4K (upsampled)0.21 USD

Source: Google AI Studio pricing page, accessed 2026-11-06

What the 1.1 price was

Gemini Omni 1.1 Flash entered preview on Aug. 27, according to eesel.ai's pricing analysis. That analysis puts 720p at about $0.101 per second, and says Google itself described it as "approximately $0.10 per second." Eesel derived its other resolutions from token counts, so only the 720p figure is Google's.

Google's pricing page says that from Dec. 3, 2026, requests to gemini-omni-1.1-flash automatically route to gemini-omni-1.2-flash and bill at 1.2 rates. Anyone with a budget built on the old rate gets the lower price without changing code.

What 15 seconds buys

Duration is set in one-second steps from 3 to 15 seconds. Google's blog lists three kinds of reference input: character, product or style images, motion clips up to 3 seconds, and audio cues. Output carries an imperceptible SynthID watermark.

Eesel reported that 1.1 reached 40 seconds only by chaining scene extensions in 10-second steps. Google's blog does not say whether scene extension carries over to 1.2.

The same clip length at other resolutions, from the per-token rate, is about $0.35 at 360p, $1.04 at 720p, $1.56 at 1080p and $3.13 at 4K. Those are this publication's multiplications.

Pricing models in video are not settled. Kino AI's agent-ready video editor charges $25 a seat, while Google meters by the second. Open-weight image generation, such as Linum's Apache 2.0 model, has no per-output meter at all.

What Google has not shown

The blog post carries no benchmark numbers and names no outside evaluator. Its evidence for better consistency is side-by-side demos of 1.1 against 1.2, including a poster reference test and a first-frame to last-frame test.

Google lists what is still weak: fast sports action, rapid script changes and complex physics. It describes the model as ideal for producing video "at scale," which is a claim about price, not quality.

Availability is staggered. Google Vids gives 15-second clips at no extra cost to anyone with a Google account or Workspace subscription. The Gemini app is rolling the feature out to Google AI Ultra subscribers. Provisioned Throughput is available on Google Cloud.

The Dec. 3 routing date is the next fixed point. After it, the 1.1 model ID stops being a separate product.