Gemini Omni 1.2: Check Your Workflow Before December 3

On November 5, 2026, Google announced Gemini Omni 1.2 Flash for video generation and editing: an ordinary clip can now run up to 15 seconds, rather than 10, and the standard video-output rate falls from $17.50 to $12 per million tokens. Google also says API requests using the older gemini-omni-1.1-flash model name will route automatically to 1.2 on December 3, 2026, at 1.2 rates. That is a reason to check a production workflow now, particularly if it uses frame interpolation or extension: those operations retain separate 10-second limits. These changes come from Google’s announcement, API changelog and pricing table; we have not run the model or measured an invoice.

What changes when 1.1 routes to 1.2

The November 5 API changelog calls Omni 1.2 Flash generally available and gives the December 3 routing notice. The pricing page confirms that requests sent to the old model name will be billed at the new model’s rates after that switch. A request that still succeeds with the old string will therefore not prove it used 1.1. Log the requested model name, returned model details where available, operation, duration and usage before comparing results across the date.

Published setting	Omni 1.1 Flash	Omni 1.2 Flash
Maximum ordinary generated or edited clip	10 seconds	15 seconds, selectable in one-second steps from 3 to 15
Standard video output, per million tokens	$17.50	$12.00
Standard text, image or video input, per million tokens	$1.50	$1.50
Standard text output, per million tokens	$9.00	$9.00
Free API tier for these models	Not available	Not available
Request to the gemini-omni-1.1-flash alias	Uses 1.1 before switch	Automatically routes to 1.2 from December 3

The duration change is stated in Google’s announcement; the token rates, free-tier status and alias behavior are on its pricing page. The prices are public standard paid-tier rates, not a quote for a particular account. Google also announces doubled paid-tier rate limits in the changelog, but a numeric quota for an individual account was not verified here. Check your own tier and limits before raising concurrency.

Google describes better reference adherence and conversational editing. Those are maker claims, not comparative results measured by DailyTech. The migration decision should depend on clips accepted by your own review standard, rather than the announcement’s examples.

Fifteen seconds does not apply to every operation

The API changelog separates ordinary generation from two other workflows. First-and-last-frame interpolation remains 3 to 10 seconds. Each video extension turn is also 3 to 10 seconds, although extensions can bring the total video to 30 seconds. A 15-second ordinary clip limit does not mean an extension call accepts 15 seconds, or that a single operation outputs 30 seconds.

Omni 1.2 supports up to six reference images according to the same changelog. That is a maximum, not proof that six references improve a specific output. Google’s Cloud article notes that audio references require allowlisting. It displays November 6 as its publication date, one day after the main November 5 announcement. Its statements about production use and Provisioned Throughput describe Google Cloud’s offering; they do not establish that every account has every feature. Verify entitlement in the account and endpoint you actually use before changing a job template.

For a pipeline that makes standard clips, the longer duration may eliminate a separate stitching step. For a pipeline driven by endpoint-specific extension or interpolation, the request-duration validator should stay tied to that operation. Neither outcome is guaranteed without running your own workload.

What the lower output rate means for a clip

Google’s pricing page says 720p video output consumes 5,792 tokens per second. At the standard 1.2 video-output rate of $12 per million tokens, an illustrative 10-second output component is 10 × 5,792 × $12 / 1,000,000 = $0.69504, about $0.70. A 15-second component is 15 × 5,792 × $12 / 1,000,000 = $1.04256, about $1.04. The same 10 seconds at the old $17.50 rate would be 10 × 5,792 × $17.50 / 1,000,000 = $1.01360 for video output alone.

The rate change works out to (17.50 − 12.00) / 17.50 = 31.43% for this line item. Google describes it as 30% lower in its announcement, a rounded marketing figure. The table rate is the figure to use in arithmetic. A longer 15-second clip has a larger output component than a 10-second clip even at the lower rate; changing duration and changing rate are separate decisions.

These are output-only calculations, not complete generation prices or observed bills. The same pricing table lists $1.50 per million input tokens and $9 per million text-output tokens. Prompts, image or video references, generated text, edits and upscaling may add billable token usage according to what you request and the usage returned. Resolution also changes tokens per second. Record the actual usage and rate for each component, then total the charged components. Our video API cost guide explains why a final accepted clip may cost more than one generation attempt. The API cost calculator can help with text-token arithmetic, but its model menu is not a verified Omni video-output quote.

What to check before the switch

Start by searching application configuration, queued jobs and saved templates for gemini-omni-1.1-flash. If a stable 1.1 output is required for a comparison or approval process, plan the comparison before December 3; the alias notice leaves no promise of continued 1.1 behavior afterward. If you move explicitly to gemini-omni-1.2-flash, treat that as a workflow change and record the model string in output metadata.

Next, sort jobs by operation rather than by model alone. Keep standard generation’s 3-to-15-second choice separate from interpolation and per-turn extension’s 3-to-10-second ranges. Inspect any client validation, timeout, polling and storage logic that assumes a fixed 10-second result. Google’s public claims do not tell you whether a longer clip will fit your own latency or quality target.

Then run a small comparison with fixed prompts, references, duration and resolution in your own account. Count billable attempts, accepted outputs and returned token usage; compare like with like. Use an actual accepted-clip cost, including unsuccessful or revised attempts if billed, before changing a budget. The image API cost guide explains the same acceptance-rate distinction for still images, although its image rates do not price Omni video. Finally, confirm account access to any gated audio-reference feature and check the current model listing in the broader AI model tracker. The tracker did not list Omni 1.2 when checked on November 7; Google’s dated pages are the evidence for this release.

Sources

• Google announcement — November 5 release, ordinary clip duration and stated product improvements.
• Gemini API changelog — November 5 GA entry, operation limits and December 3 alias routing.
• Gemini API pricing — 1.1 and 1.2 rates, free-tier status and 720p output-token conversion.
• Google Cloud article — November 6 Cloud availability notes and audio-reference allowlisting.

All pages read on November 7, 2026.

Free toolWhat will 10,000 chats a month cost?Estimate my bill