Google released Gemini Omni 1.2 Flash on Nov. 5, 2026, and said the video-generation model is generally available on the Gemini API and Gemini for business^1,2. Nobody independent has scored it yet on the tests Epoch AI and other third parties use for the models already on the scoreboard⁵. Google Cloud lists 720p generation at $0.07 per second, 30% below Gemini Omni 1.1 Flash, and OpenRouter lists token rates of $1.50 per million input tokens and $9 per million output tokens, with a 131,072-token context window^2,6. Developers can use the paid Gemini API and Google AI Studio; businesses can use Google Cloud, Google Vids and related apps; Google AI Ultra subscribers are in a Gemini-app rollout^1,2,4.

The scoreboard still has no independent result for Gemini Omni 1.2 Flash on Epoch AI's Capabilities Index, a single number Epoch AI builds from many tests so models that sat different exams can be compared⁵. It also has no independent score on GPQA Diamond, graduate-level science questions; FrontierMath tiers 1-3, unpublished math problems; ARC-AGI-2, visual puzzles that need a new rule each time; SciCode, research-style coding; APEX-Agents, one-shot professional tasks; SimpleQA Verified, short closed-book facts; or CritPt, unpublished physics problems⁵.

Those tests have been run on other recent models. Epoch AI puts Anthropic's Claude Fable 5.2, released Nov. 9, at 173.4 on the Capabilities Index, Google DeepMind's Gemini 4 Carbon, released Nov. 10, at 169.8, and OpenAI's GPT-6 Astra, released Sept. 3, at 169⁵. On GPQA Diamond, Epoch AI has Gemini 4 Carbon at 96.5%, GPT-6 Astra at 95.8% and Claude Fable 5.2 at 93.9%⁵. On FrontierMath tiers 1-3, Epoch AI has Claude Fable 5.2 at 95.1%, GPT-6 Astra at 93.7% and Gemini 4 Carbon at 90.5%⁵. None of those figures apply to Gemini Omni 1.2 Flash⁵.

Google presents Omni 1.2 Flash as a video model, not as a drop-in rival on those text and agent tests^1,2. Google says native clips now run up to 15 seconds, up from 10 seconds on Omni 1.1, with duration set in one-second steps from 3 to 15^1,3. Google says users can edit the same clip over several chat turns instead of regenerating it, and that outputs carry a SynthID watermark^1,2. Google DeepMind's own side-by-side tests, reported by Unite.AI and The Next Web, give Omni 1.2 a 62% visual-consistency win rate against 38% for Omni 1.1 in text-to-video, and a 63% multi-reference identity-consistency win rate against 37%^3,4. Those numbers are Google's, not an outside lab's^3,4.

Unite.AI, citing Google's developer pricing, says paid Gemini API video output is $12 per million tokens, metered at 5,792 tokens per second of 720p, or about $0.70 for 10 seconds and $1.04 for 15 seconds, and that the model is not on the free tier³. Google says it is still working on fast sports action, rapid script changes and complex physics^1,4. There is no previous Google text model in the assignment to compare on independent scores, and the scoreboard does not rank Gemini Omni 1.2 Flash among the 44 ranked models⁵.