A 15-second 720p clip comes to about $1.04 on the official pricing page. Swap the product, dim the lighting, move the camera, and you do not rebuild the whole clip; you say it in a conversation. This is the generally available (GA) entry in the Gemini API official changelog on November 5, 2026: model code gemini-omni-1.2-flash, native video raised from the 10 seconds of version 1.1 to 15 seconds, video output pricing cut 30 percent to $12.00 per 1M tokens, and paid-tier rate limits doubled. It is not a brand-new model: 1.1 was already listed, and on November 5, 2026 the 1.2 release reached GA rather than launching from zero.
Scope first, so angles do not collide. This piece covers the GA event of November 5, 2026 plus the capability boundaries and pricing Google published alongside it. In the same batch, the open-source column: the craft-anything project takes apart that single repository, the comparison column: five GitHub tools side by side lines up objective parameters for five tools, and the SOP column: a five-step API integration and cost math owns the five-step integration and the cost table; the four do not overlap. Two earlier posts here, the AI drama tools comparison of November 2, 2026 and the Flux versus Seedance comparison of October 22, 2026, answer a different question: which tool to pick. This article is no tool-selection exercise; it is about one model, gemini-omni-1.2-flash, reaching GA and about the official billing rates.
November 5, 2026 Is a GA Date, Not a Launch Date
Pin the dates down first. The Gemini API official changelog entry for November 5, 2026 reads generally available (GA), Released Gemini Omni 1.2 Flash (gemini-omni-1.2-flash), open on the paid tier of the Gemini API. The Google Cloud official blog followed on November 6, 2026, and the Chinese-language coverage on November 6, 2026 was reprints; neither is the release date. The date to remember is November 5, 2026, and the label is GA, generally available.
GA does not mean a new model appearing out of nowhere. The official model page lists both gemini-omni-1.2-flash and gemini-omni-1.1-flash under Stable, marks gemini-omni-flash-preview as Shut down, and gives Latest update as November 2026. Version 1.1 was already stable; 1.2 pushes the line to general availability. Reading it as a launch from scratch would erase the 10 seconds ceiling and the $17.50 video-output price of 1.1. Veo 3.1 Preview in the Gemini API was shut down on October 22, 2026, a separate matter.
The same changelog paragraph stacks the change set into one entry: generation and conversational editing up to 15 seconds, first and last frame interpolation at 3 to 10 seconds, per-turn extension at 3 to 10 seconds up to 30 seconds total, consistency across up to six reference images, video output pricing 30% lower at $12.00 per 1M tokens, and doubled paid-tier rate limits.
From 10 Seconds to 15, and the Edit You Do Not Redo
Version 1.1 capped native video at 10 seconds; 1.2 raises that to 15 seconds, with a clip selectable from 3 to 15 seconds and 10 seconds still the default. The official video generation document is blunt: a duration over 15 seconds returns an invalid argument error.
Extension rules matter more than the single-clip number. First and last frame interpolation runs 3 to 10 seconds; supplying both frames with a duration above 10 seconds errors out too. Video extension runs 3 to 10 seconds per turn, up to 30 seconds total, tail-only, with no prepending or middle insertion. Reference videos have their own slicing: a 3 to 10 second clip supports up to 3 reference videos, an 11 to 15 second clip supports 1, each at most 3 seconds. Longer therefore means appendable tail growth across turns, not one continuous thirty-second take.
Conversational editing is what the headline calls editing becoming a chat. The Google Cloud official blog describes refining the same clip repeatedly: swap a product, change the lighting, shift the camera angle, without redoing the whole thing. The official Omni guide files it under multi-turn conversational editing, with previous_interaction_id tracking video state so later turns need no re-upload. Editing prompts stay short and end with "Keep everything else the same"; negatives go into the regular prompt, for example "No dialogue" or "No embellishments"; "In a single unbroken scene" or "No scene cuts" keeps one clip from splitting into shots.
Consistency rests on up to six reference images, up to 3 per character for up to 2 characters. Output is video with native audio, with more fluid dialogue, clearer lip-sync and distinct multi-speaker conditioning.
30 Percent Cheaper: Reading the Bill Down to One Second
Copy prices from the official pricing page and never mix them with 1.1. The Standard paid tier for Gemini Omni 1.2 Flash reads: Free Tier marked Not available, input $1.50 per 1M tokens (text, image and video), text output $9.00, video output $12.00. Against the $17.50 video-output price of 1.1, the $12.00 of 1.2 is exactly the 30 percent cut the changelog names; the page gives 1.1 an effective price of about $0.10 per second.
Billing is based on total output token consumption, not a flat fee per clip. The official pricing page sets 5,792 tokens per second of 720p video; at $12.00 per 1M tokens that is about $0.07 per second, calculated from the official unit price. It then states two figures directly: about $0.70 for a 10s clip and about $1.04 for a 15s clip. The $1.04 deserves care: it is the number printed on the official pricing page, not a value to recompute by multiplying a per-second rate, and it should not be swapped for another number. The official token conversion and a rough per-second multiplication do not line up exactly, so quoting the page verbatim is safest.
Copy the resolution tiers too: 360p is 1,931 tokens per second, about $0.023 per second; 1080p with upsampling is 8,688 tokens per second, about $0.104 per second; 4K with upsampling is 17,376 tokens per second, about $0.209 per second. 360p and 720p are drafts, 1080p and 4K are upscaled tiers, and the blog suggests drafting at 360p or 720p before upscaling to 1080p or 4K. One more line, "Used to improve our products", is Yes on the Free Tier and No on the paid tier.
Two Channels, Two Sets of Rules, No Cross-Explaining
Keep Gemini API and Google Cloud separate. On the Gemini API paid tier side: Free Tier is Not available, so there is no free tier; the official video generation document states "Provisioned throughput is not supported", so Provisioned Throughput is not supported here. The document also lists system instructions, temperature, top_p, stop sequences and negative prompts as unsupported, with negatives folded into the regular prompt; uploading audio references is unsupported in the current version of the API; voice editing and YouTube as a media source are not supported; English is fully supported while other languages have not been evaluated; reference videos are capped at 3 clips.
The Google Cloud and Vertex AI side is the November 6, 2026 official blog: it calls Gemini Omni 1.2 Flash production-ready, backed by enterprise-grade infrastructure and security, and offering Provisioned Throughput. Audio references are gated behind allowlisting. Outputs carry an imperceptible SynthID watermark. The blog repeats $0.07 per second of 720p video, 30 percent less than 1.1 Flash, clips up to 15 seconds up from 10s, 360p or 720p drafts upscaled to 1080p or 4K, clearer lip-sync, and editing in conversation that swaps a product or changes lighting or camera angle.
The two must not explain each other. The Cloud offering Provisioned Throughput cannot be poured into the API's "Provisioned throughput is not supported" line; production-ready on Cloud does not imply a free tier on the Gemini API. Audio references too: the API in its current version does not support uploading them, while the Cloud side opens them through allowlisting, one statement per channel. GA addresses the paid tier of the Gemini API; the Cloud blog describes a different channel.
December 3 Has Not Arrived: Routing Is a Future Change
The changelog and pricing page carry a dated notice: "Effective December 3, 2026, API requests to gemini-omni-1.1-flash will automatically route to gemini-omni-1.2-flash at Gemini Omni 1.2 Flash rates." From December 3, 2026, requests pointed at 1.1 route to 1.2 automatically and are billed at 1.2 rates.
The effective date sits in the future. As of November 7, 2026 this rule is a future change that has not taken effect and must not be reported as active. Nor does 1.1 disappear: the model page still lists it under Stable with a note "routes to 1.2 on Dec 3, 2026". For developers, old code carrying the 1.1 model code still gets a response after December 3, but 1.2 runs behind it, and the bill follows the $12.00 video-output price of 1.2 rather than the $17.50 of 1.1. The step-by-step integration and cost table belong to the SOP column; the hotspot column only pins the effective date.
The Hard Numbers on the Model Page: 131,072 and 65,536
The official model page gives gemini-omni-1.2-flash inputs of text, image and video (up to 15s for editing and 10s lookback for extension) and outputs of video with native audio plus text; the input token limit is 131,072 tokens and the output token limit 65,536 tokens. Output video runs 3s to 15s at 360p, 720p, 1080p or 4K at 24 FPS, with interpolation and per-turn extension at 3 to 10 seconds up to 30 seconds total.
The video generation document spells out the parameters: response_format takes type "video", aspect_ratio can be "9:16" (default "16:9"), resolution takes 360p, 720p (default), 1080p (upscaled) or 4k (upscaled), duration takes "3s" to "15s" (default "10s"), and videos over 4MB use delivery="uri". A call reading only the official document looks roughly like this:
interaction = client.interactions.create(
model="gemini-omni-1.2-flash",
input="A single unbroken scene, no scene cuts",
response_format={"type": "video", "resolution": "720p", "duration": "10s"},
)Every generated video includes SynthID watermarking, invisible and programmatically detectable, and content safety filters apply to input prompts and generated video. Clips over 4MB go through delivery="uri", Files API uploads are polled until ACTIVE, and background=false, store=false and stream=false speed up synchronous generation, while store=false makes the clip non-editable later. Where official pages still carry 1.1 examples they should not be followed; the model code is always gemini-omni-1.2-flash.
Who Should Start Now, and a Few Cost-Saving Habits
Draft at 360p or 720p first, use six reference images to lock characters and subjects, then iterate with conversational editing across turns, leaning on previous_interaction_id to keep state without re-uploading. Once settled, upscale the chosen clip to 1080p or 4K.
The cost math follows the same order. A 15-second 720p clip is about $1.04 on the official pricing page, 360p is $0.023 per second and 720p is $0.07 per second; calculated from the official unit price, the 360p per-second figure is roughly one third of the 720p figure, so polishing edits on a low-resolution draft before upscaling costs less than regenerating at high resolution from the start. For a service integration: clips over 4MB go through delivery="uri", Files API uploads are polled until ACTIVE, editing prompts stay short and end with "Keep everything else the same", and negatives go into the regular prompt. The five-step sequence and cost table live in the SOP column.
One note of restraint. Google publishes a capability boundary and a price list, not a guarantee of results in any given scenario, and bills follow each clip's own token rates. This article copies only the numbers and limits on the official pages and adds no self-run conclusions.
Keeping the Dates and the Numbers in Their Places
To close. On November 5, 2026 the Gemini API official changelog marked gemini-omni-1.2-flash as generally available on the paid tier: a clip rose from the 10 seconds of 1.1 to 15 seconds, with 10 seconds still the default and anything over 15 seconds returning an invalid argument error; video output fell from $17.50 to $12.00, a 30 percent cut; a 15-second 720p clip is about $1.04 on the official pricing page, off a baseline of 5,792 tokens per second and about $0.07 per second. Conversational editing, up to six reference images, native audio, the SynthID watermark, the 131,072 and 65,536 token limits and 24 FPS sit in the same set of official pages.
Keep three dates apart: November 5, 2026 is the GA date, November 6, 2026 is the Cloud blog follow-up and Chinese reprint date, and December 3, 2026 is the future effective date on which 1.1 requests route to 1.2 at 1.2 rates, a change that as of November 7, 2026 has not taken effect. Keep two channels apart: the Gemini API paid tier has no free tier and its document says "Provisioned throughput is not supported", while Google Cloud calls the model production-ready, offers Provisioned Throughput and gates audio references behind allowlisting. Treat GA as the event, the official pricing page as the price authority and the changelog as the capability list, and none of these lines will drift.
Related reading