Google's Gemini Omni Flash Dominates All Three Video Leaderboards at Once

Artificial Analysis announced that Google's Gemini Omni Flash debuted at #1 on their Video Editing Leaderboard, completing what they called a near-sweep of all three video leaderboards. The model had already held the top spot in Text-to-Video and Image-to-Video, making it the leader across every major video category simultaneously. Since that announcement, MiniMax H3 has entered the leaderboard and now sits at #1 in Video Editing, pushing Gemini Omni Flash to #2.

Where things stand now

The current Artificial Analysis Video Leaderboard rankings show a tight race across all three categories:

The margins are tight enough to fall within statistical error in most cases. Rankings use an Elo system derived from millions of blind pairwise votes, where users pick between two unlabeled videos. The Video Editing Leaderboard currently has between 8,000 and 18,000 samples per model, giving the scores statistical weight.

What Gemini Omni Flash actually is

Gemini Omni Flash (gemini-omni-flash-preview) is a multimodal model built for high-speed video generation, editing, and cinematic control. The "omni" in the name refers to its architecture: it processes video, audio, text, and images in a single unified context rather than routing each modality through separate pipelines.

The feature that sets it apart from earlier video models is conversational editing. Through the Interactions API, you can refine a video across multiple turns of natural language, describing what to change while the model preserves everything else. Each edit builds on the previous one, so you never re-prompt from scratch.

A two-turn editing session in Python looks like this:

from google import genai
import base64
client = genai.Client()
# Turn 1: Generate initial video
res1 = client.interactions.create(
    model="gemini-omni-flash-preview",
    input="A woman playing violin outdoors."
)
# Turn 2: Edit it, model preserves 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."
)
with open("edited.mp4", "wb") as f:
    f.write(base64.b64decode(res2.output_video.data))

Where it leads, and where it falls short

Artificial Analysis grouped video editing prompts into five capability areas. Gemini Omni Flash tops four of them:

• Complex Edits: stacking multiple instructions into a single request, where it finishes 117 Elo points clear of the next model
• Object Editing: adding, removing, or replacing objects in a scene
• Visual Effects Editing: style transfers, lighting changes, and on-screen text corrections
• Physics and World Simulation: edits that require understanding how real-world objects behave

Audio is the exception. Despite generating native audio alongside video, Gemini Omni Flash places third in Sound and Speech Editing, behind Alibaba's Wan 2.7 and HappyHorse 1.0, and fifth in SFX. Unified multimodal processing produces more cohesive output overall, but it does not automatically produce best-in-class audio editing.

One other caveat: Artificial Analysis noted that Gemini Omni Flash declined to generate some prompts due to content filters, most commonly on prompts involving people. Its leaderboard scores reflect only the prompts it successfully completed.

Pricing

The model costs $0.10 per second of video output, the same rate as Veo 3.1 Fast, which works out to $6.00 per minute. Editing is billed at the same rate as generation with no separate surcharge. For comparison, HappyHorse 1.0 (currently #3 on the Video Editing leaderboard) costs $27.04/min, and Runway Aleph 2.0 runs $16.80/min. Gemini Omni Flash is among the cheapest top-tier options available.

How to access it

Gemini Omni Flash is available to developers through the Gemini API and Google AI Studio, accepting text, image, and video inputs for both generation and conversational editing. It also runs on the Gemini Enterprise Agent Platform and on consumer surfaces including the Gemini app, Google Flow, and YouTube Shorts.

The model ID is gemini-omni-flash-preview. You can upload your own videos for editing via the Files API, though regional restrictions apply: users in the EEA, Switzerland, and the UK can edit model-generated videos but not uploaded ones.

One model across three leaderboards

Most video models are purpose-built: generation-only, editing-only, or strong at a single input modality. A single model sitting near the top across text-to-video, image-to-video, and video editing simultaneously is unusual. Gemini Omni Flash gets there through multi-turn conversational editing, multi-input reasoning across text, images, audio, and video, and strong character and scene consistency.

MiniMax H3, an open-weights model that now leads the Video Editing leaderboard and is statistically tied with Gemini Omni Flash in Text-to-Video, shows the competition is real and closing fast. Even so, the combination of top-tier quality, conversational editing, and $6/min pricing makes Gemini Omni Flash a strong default for developers building video workflows on an API today.