Google to reroute Gemini Omni 1.1 Flash API calls to 30% cheaper 1.2 on December 3

A cheaper model, shipped on November 5

Google made Gemini Omni 1.2 Flash generally available in the Gemini API on November 5, cutting the cost of its video generation model by roughly 30%. According to a dev.to write-up citing Google's pricing page, output video now bills at $12.00 per million tokens, down from $17.50 per million for Omni 1.1 Flash — about $0.07 per second of 720p video, against roughly $0.10 before.

The new model also does more. Per the launch post as summarized on dev.to, a single generation can run up to 15 seconds, up from 10, set in one-second steps between 3 and 15. It supports multi-turn editing — swapping a product, background, lighting or aspect ratio without regenerating the whole clip — and improves consistency for faces, garments, typography and brand colors, with up to six reference images. Drafts render at 360p or 720p and can be upscaled to 1080p or 4K, and every output carries an invisible SynthID watermark. Google itself flags weak spots in fast sports action, rapid script changes and complex physics.

Google DeepMind's model page claims a 62% to 38% win rate for 1.2 over 1.1 on visual consistency, a figure The Next Web has reported as testers preferring the newer model. As the dev.to author notes, it is a vendor's number, and it leaves room for plenty of clips where 1.1 looked just as good.

The December 3 reroute

The change that reaches existing deployments sits in the changelog. From December 3, 2026, requests to gemini-omni-1.1-flash will automatically route to gemini-omni-1.2-flash and bill at 1.2 rates. Google's deprecations page, quoted by dev.to, states that the endpoint "is automatically routing API requests to gemini-omni-1.2-flash", and the Omni documentation adds that no code changes are required. No commit, config edit or deployment will record the day the model behind the identifier switched.

The cut applies per second, not per call. A 15-second 720p clip on 1.2 costs about $1.04, a figure Google publishes, while a 10-second 1.1 clip at the same resolution works out to about $1.01 by the dev.to author's arithmetic on token rates. Products that adopt the longer maximum length will therefore see slightly higher per-call bills despite the cheaper rate, and a 15-second upscaled 4K clip runs to roughly $3.13.

Neither model has a free tier. A paid Tier 1 project is limited to 8 requests per minute and 40 per day, while Tier 3 gets 40 per minute and 2,000 per day, with limits applying per project rather than per API key. The changelog notes that paid-tier limits doubled with this release.

Behavior changes behind the same ID

The reroute is not purely financial. Google's model page lists 1.1's output as video only, whereas 1.2 produces video with native audio and text, so any pipeline step that assumes a silent track needs review. For editing and extension, 1.1 accepted input video up to 10 seconds; 1.2 takes up to 15 seconds for edits, applies a 10-second lookback when extending, and can carry a clip to 30 seconds total.

The dev.to post also highlights two gaps between the launch post and the API documentation. The announcement names "audio cues" among reference inputs, but the docs say uploading audio inputs or audio references is unsupported, and audio inside a reference video is ignored; the launch post also covers consumer products like Flow and the Gemini app, where the capability may actually live. Separately, multi-turn editing depends on stored interactions: setting store=false for data-handling reasons means a generated video cannot be edited in later turns, so every tweak becomes a full-price regeneration.

What to do before the switch

The dev.to author recommends searching codebases for hardcoded model IDs, including the older gemini-omni-flash-preview, which the deprecations page lists as shut down since October 22. Migrating to gemini-omni-1.2-flash behind a configuration value lets a team pick its own cutover date, and re-running a fixed set of prompts and reference images against 1.2 — then comparing audio, duration and file size against shipped 1.1 output — surfaces regressions before December 3 forces the change. Moving early also secures the lower price immediately.

Why it matters

When a provider swaps the model that answers a pinned identifier, applications receive a quality and behavior change that nobody on the team shipped. Here the shift is mostly favorable — cheaper output, longer clips, native audio — but the mechanism needs no approval, leaves no trace in version control, and is becoming routine: the same dev.to post notes a December 1 reroute of gemini-2.5-pro calls to Gemini 3.8 Flash on identical terms. Teams building on managed model APIs are left with a few defenses: log which model actually served each request, keep an evaluation set that can be re-run the week a vendor swaps models, and track cost per request rather than monthly totals.