Gemini Omni 1.2 Flash Is Generally Available With 15-Second Clips, and Google Cloud Lists Video Output at $12 per Million Tokens, Down From $17.50

Overview

Google has made Gemini Omni 1.2 Flash, its video generation and editing model, generally available. In a post dated November 5, 2026, Google says it is “generally available across the Gemini API and Gemini for business”, that clips can now run up to 15 seconds, up from 10 seconds with Omni 1.1, and that it has “dropped the price 30%” to $0.07 per second for 720p video. The Google Cloud blog is dated November 6 and says the model is “generally available on Google Cloud”, while Google Cloud’s model documentation lists a release date of November 5, 2026. Dates in this article are as printed on each of these three pages.

Every capability claim below is Google’s unless another source is named. The Decoder and The Next Web report the launch based on Google’s statements. The independent evaluator Artificial Analysis, on three of its video leaderboards as the Herald read them on November 9, lists Omni Flash 1.1 and no 1.2 entry.

What Google Says Changed

According to Google’s post, duration can be set in one-second increments from 3 to 15 seconds, multi-turn conversational editing works across full 15-second clips, and users can guide generation with character, product or style images, short motion clips of up to three seconds, and audio cues. It also describes a workflow of drafting at 360p or 720p and upscaling the best results to 1080p or 4K. The post says Provisioned Throughput is available and that outputs carry an imperceptible SynthID watermark.

Google names remaining weak spots: it says development is ongoing on “fast sports action, rapid script changes and complex physics”. The post’s comparison of 1.1 and 1.2 consists of Google’s own example prompts and captions; it publishes no preference study or benchmark table.

For consumers, the same post says 15-second clips come to Google Vids at no additional cost and to the Gemini app for Google AI Ultra subscribers on a rolling basis. The Next Web notes that ByteDance’s Seedance 2.5 makes 30-second clips.

Developer Details on Google Cloud

Google Cloud’s model documentation gives the model ID gemini-omni-1.2-flash, a launch stage of GA, and a retirement date of “November 5, 2027 or later”. It lists the Interactions API as supported and the GenerateContent API as not supported. It also lists Provisioned Throughput as supported and, among other features, batch inference and context caching as not supported. Output runs 3 to 15 seconds for generation and editing, with 3 to 10 seconds for first-and-last-frame video and for extension; supported aspect ratios are 16:9 and 9:16.

Google’s pages differ on audio input. The model documentation lists audio under modalities as “Not supported”, although the pricing page’s input row names audio and states 32 tokens per audio second. Google’s post lists audio cues among the inputs for guiding generation, and the Google Cloud blog says “Audio references are gated behind allowlisting.” None of the Google pages cited here explains the difference, so it is reported here without reconciliation.

Pricing

Google Cloud’s pricing page lists Omni 1.2 Flash at $1.50 per million input tokens, $9.00 for text output and $12.00 for video output. The row for “Gemini Omni Flash, Gemini Omni 1.1 Flash” lists the same input and text prices and $17.50 for video output. The page says video output is charged at 1,931 tokens per second at 360p, 5,792 at 720p, 8,688 at 1080p and 17,376 at 4k.

The following figures are the Herald’s own arithmetic on those listed prices. At 720p, 5,792 tokens at $12.00 per million is about $0.0695 per second, in line with Google’s $0.07; at $17.50 it is about $0.1014, in line with the $0.10 per second that The Next Web gives for the earlier price. The per-token cut from $17.50 to $12.00 is 31.4%, while Google states 30%; the rounded per-second figures ($0.10 to $0.07) also give 30%. At 1080p the new rate works out to about $0.104 per second, or $6.26 per minute, against $9.12 per minute at the old rate. A 15-second clip at 720p would cost about $1.04.

Independent Measurements

Artificial Analysis ranks video models by “Elo from pairwise human preference votes”. On the Herald’s reading on November 9, its text-to-video board places Gemini Omni Flash 1.1 at rank 12 (range 9-12) with an Elo of 1128 (±13) from 7,932 samples. Its image-to-video board places it at rank 16 (range 15-17) with 1044 (±14) from 6,478 samples, and its video-editing board at rank 4 (range 2-4) with 1146 (±10) from 8,271 samples. The text-to-video board lists the 1.1 price as $9.12 per minute and states that its prices reflect API cost “at the settings we benchmark (1080p where offered)”; the Herald’s 1080p calculation above arrives at the same figure.

None of those three pages lists a 1.2 entry, so the Herald found no independent score for the new version, and the reported gains in reference adherence, consistency, dialogue and camera control rest on Google’s description.

What Is Not Known

The Next Web reports that Google DeepMind’s own comparisons found testers preferred 1.2 over 1.1 for visual consistency 62% of the time. That figure does not appear in Google’s post, the Google Cloud blog, the model documentation, the pricing page or the model card, and the Herald did not find its source.

Google’s model card says 1.2 Flash “is based on Gemini Omni Flash” and refers readers to that model’s card for architecture, known limitations and evaluation approach; the card carries no benchmark scores. It also states that the model can change people’s speech when editing videos and that “For now, we are restricting this capability”.