Google will route Omni 1.1 requests to 1.2 Flash. Check the transition date, output-only cost of a 15-second clip and your acceptance criteria.
On December 3, 2026, requests to gemini-omni-1.1-flash will automatically route to gemini-omni-1.2-flash, according to Google's API documentation. Code using the old identifier can keep sending requests while a different model version serves them. As of November 6, this is a scheduled transition, not a switch that has already happened. A production video workflow should retain prompts and reference outputs before the change so that the replacement can be compared against them.
In the November 5 announcement, Google DeepMind product manager Anish Nangia introduced clips up to 15 seconds, up from 10, and editing through successive conversational turns. He also claims better adherence to references. That is a vendor claim: NODVIEW has not generated videos or tested faces, packaging or lettering. Google's own announcement says fast sports action and complex physics still need improvement.
For budgeting, use the API pricing details, not just the headline price per second. Google lists $12 per million output video tokens and 5,792 tokens per second of 720p video. An illustrative 15-second output therefore costs 5,792 × 15 × 12 / 1,000,000 = $1.04256, or about $1.04 for video output alone. This is our calculation from published rates, not a test invoice: inputs are charged separately, while retries and upsampling change the total. This model has no free Gemini API tier.
Before migration, assemble a small set of your own scenes using references you have the right to use. Define acceptance criteria for readable text, product details, motion and aspect ratio. Retain the requested model identifier, date, duration, resolution and actual spending across all attempts. After automatic routing starts, the old identifier alone will no longer establish that version 1.1 ran. A lower rate does not establish a lower cost per accepted clip if your scene needs more attempts; evaluate that on your own set rather than extrapolating from Google's demonstrations.