---
title: "Google's Gemini Omni Flash Tops Video Leaderboards With Built-In Reasoning"
subtitle: "Google's Gemini Omni Flash debuts at the top of Artificial Analysis video leaderboards, beating Seedance 2.0 with conversational editing at $6/min"
url: "https://alphasignal.ai/news/google-s-gemini-omni-flash-tops-video-leaderboards-with-built-in-reasoning"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "Artificial Analysis"
company_url: "https://alphasignal.ai/company/artificialanlys"
published: "2026-07-13T19:14:21.000Z"
updated: "2026-07-13T20:19:30.615Z"
type: "News"
topics:
  - "Video"
  - "Benchmarks"
subtopics:
  - "Video Generation"
upvotes: 774
read_time: "6 min"
source_url: "https://artificialanalysis.ai/video/leaderboard/image-to-video"
comments: 0
---

# Google's Gemini Omni Flash Tops Video Leaderboards With Built-In Reasoning

> Google's Gemini Omni Flash debuts at the top of Artificial Analysis video leaderboards, beating Seedance 2.0 with conversational editing at $6/min

[Gemini Omni Flash](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni/) just landed at the top of both the Text-to-Video and Image-to-Video leaderboards on Artificial Analysis, narrowly edging out ByteDance's Seedance 2.0 on each. But the ranking is almost secondary to what the model actually represents: Google's first serious attempt at fusing its Gemini reasoning engine directly into a video generation model, rather than keeping the two as separate systems bolted together.

## Not just another video generator

Gemini Omni Flash is Google's next step toward models that can create and edit anything from any input, starting with video. It combines Gemini's intelligence with Google's generative media models, representing a leap forward in world understanding, multimodality, and editing.

The key architectural distinction between Gemini Omni Flash and Veo 3.1 is not resolution or speed -- it is world understanding. Veo 3.1 is a specialized video generation model optimized for high-fidelity output from text prompts and reference images. Gemini Omni Flash is built on Gemini's unified transformer architecture, which means it has Gemini's multimodal reasoning, world knowledge, and contextual understanding built in, not bolted on.

In practice, this means the model can do things like accurately simulate the water cycle with labeled arrows, render optical physics phenomena like total internal reflection in an aquarium, or produce a claymation explainer of protein folding -- all from a single text prompt.
Omni draws on Gemini's knowledge of history, biology, and narrative logic to construct compelling videos.

## The conversational editing loop

Most AI video tools today follow a generate-then-export pattern: you write a prompt, the model renders a clip, and if you want to change anything, you start over.
Gemini Omni Flash is designed for fast, conversational video generation and editing. It excels at turning text and images into video and lets you refine and edit generated videos through natural language conversations using the Interactions API.

The Interactions API tracks session state using a `previous_interaction_id`, so each edit turn builds on the last without re-uploading the video. Here is a minimal example:

```
from google import genai
client = genai.Client()
# Turn 1: Generate
res1 = client.interactions.create(
    model="gemini-omni-flash-preview",
    input="A woman playing violin outdoors."
)
# Turn 2: Edit, preserving everything else
res2 = client.interactions.create(
    model="gemini-omni-flash-preview",
    previous_interaction_id=res1.id,
    input="Make the violin invisible. Keep everything else the same."
)
```

The Interactions API currently supports stacking up to three sequential edits while maintaining session context. Beyond three edits in a sequence, context may degrade. This is a deployment configuration rather than a model limit, per Google's developer documentation, and is expected to increase over time.

## What it accepts, what it outputs

Gemini Omni Flash accepts text descriptions, images (up to seven images as references), and video clips (up to three clips of three seconds each or less) as inputs, or any combination of these. From these inputs it generates 3 to 10 second clips at 720p resolution in landscape (16:9) or portrait (9:16) aspect ratio, with native audio generated alongside the video. Every video Gemini Omni Flash produces includes synchronized audio generated natively rather than added as a post-processing step. This means the model reasons about what sound belongs in the scene based on the visual content, camera movements, and physical environment depicted.

The [official docs](https://ai.google.dev/gemini-api/docs/omni) expose four explicit task modes via the `video_config.task` parameter:

- `text_to_video`
- `image_to_video`
- `reference_to_video` -- generate from multiple reference images using `<IMAGE_REF_N>` tags
- `edit` -- modify an uploaded or previously generated video

All videos created with Omni include Google's imperceptible SynthID digital watermark. You can verify that videos were generated with Gemini Omni through the Gemini app, Gemini in Chrome, and Google Search.

## Where it falls short

Google is transparent about the current limitations.
While Gemini Omni Flash demonstrates strong progress, maintaining complete consistency throughout edits, generating scenes with complex motion, or rendering perfectly accurate text remains a challenge.
There are also several hard API constraints worth knowing before you build:

- No voice editing (capability exists but is restricted for safety review)
- No video extension or interpolation between a first and last frame
- No system instructions, temperature, or `top_p` controls
- Uploading audio references is not yet supported
- Editing uploaded videos is unavailable in the EEA, Switzerland, and the UK
- Only English has been formally evaluated; other languages may work but results vary

## Architecture and training

Gemini Omni Flash is a transformer-based model with native multimodal support for text, vision, video, and audio inputs. It was trained on audio, video, image, and text data. Audio and video datasets were annotated with text captions at different levels of detail. Training videos were filtered for compliance, safety, and quality metrics and deduplicated semantically.
Training ran on Google TPUs using JAX and ML Pathways.

Google has said image and audio output modalities are on the Gemini Omni roadmap but are not part of the initial launch.
The "starting with video" framing in the announcement is deliberate -- Omni is positioned as a platform, not a single-purpose tool.

## Pricing and availability

The model is available now as `gemini-omni-flash-preview` in the Gemini API and Google AI Studio.
This model is priced competitively at $0.10 per second of video output, which is the same as Veo 3.1 Fast.
That works out to $6.00 per minute -- the same rate for both text-to-video and image-to-video.

For context, here is how that sits against the competitive field on the [Artificial Analysis leaderboard](https://artificialanalysis.ai/video/leaderboard/text-to-video):

| Model | Creator | API Price / min |
| --- | --- | --- |
| Gemini Omni Flash | Google | $6.00 |
| Veo 3.1 Fast | Google | $9.00 |
| Seedance 2.0 | ByteDance | $9.07 |
| Veo 3.1 | Google | $24.00 |
| Kling 3.0 Pro | KlingAI | $20.16 |

Gemini Omni Flash is rolling out to all Google AI Plus, Pro, and Ultra subscribers globally through the Gemini app and Google Flow. It is also rolling out at no cost to users on YouTube Shorts and YouTube Create App.

## Where this fits in your stack

The clearest use cases are workflows that need iteration, not just generation.
For developers building product demos, marketing tools, or content automation pipelines, this dramatically compresses the iteration loop.
If your workflow involves editing real footage, analyzing content, or working across text and visual assets simultaneously, Gemini Omni Flash offers more flexibility. For teams already invested in Google's ecosystem -- using Vertex AI, Google Cloud, or other Gemini models -- Omni Flash also integrates more cleanly.

Inside the Gemini app, Gemini Omni Flash has replaced Veo as the default video generation backend for Plus, Pro, and Ultra subscribers. Veo 3.1 remains available for developers through the Gemini API and Vertex AI, and it is not going away for production use. But the consumer video generation experience is now Omni Flash.
If you need maximum photorealism and longer clips, Veo 3.1 is still the right call. If you need multi-turn editing, mixed-modality inputs, and tighter integration with the rest of the Gemini stack, Omni Flash is the new default.

## Sources

- [X post by @artificialanlys](https://x.com/artificialanlys/status/2076747075036045645)
- [Introducing Gemini Omni - Google Blog](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni/)
- [Generate and edit videos with Gemini Omni Flash - Google AI for Developers](https://ai.google.dev/gemini-api/docs/omni)
- [Gemini Omni Flash - Model Card - Google DeepMind](https://deepmind.google/models/model-cards/gemini-omni-flash/)
- [Text to Video Leaderboard - Artificial Analysis](https://artificialanalysis.ai/video/leaderboard/text-to-video)
- [Start building with Nano Banana 2 Lite and Gemini Omni Flash - Google Blog](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-flash-nano-banana-2-lite/)