Is Gemini Omni 1.1 Flash Actually Number One? What the Boards Say

Every "Google's new video model is number one" post you read today is citing a ranking for a different model.

Gemini Omni 1.1 Flash went generally available yesterday. Artificial Analysis, the board most of those posts point at, does not track it — not on text-to-video, not on image-to-video, not on video editing. The placings being quoted belong to Gemini Omni Flash, the June preview model, and most of them belong to one particular view of one particular board.

That is not a reason to dismiss the model. It is a reason to look at what has actually been measured, which turns out to be more interesting than the headline.

Where 1.1 has genuinely been ranked

One place: arena.ai, which added gemini-omni-1.1-flash to two of its boards on 27 August 2026, the day it shipped. Read on 28 August:

So yes — there is a board where Gemini Omni 1.1 Flash sits first. Now the part the headlines leave out. Second place on that board is Gemini Omni Flash at 1512 ±10, on 19,830 votes. Three Elo points, with confidence intervals that overlap completely, and roughly a tenth of the sample.

That is a statistical tie, not a win. A three-point gap on 1,762 votes is well inside the noise, and the number will move as votes accumulate. If you see this cited next month as "Omni 1.1 is the top text-to-video model", check whether the person citing it looked at the interval.

Then there is what is absent. 1.1 has not been added to any video editing board, on arena.ai or anywhere else — and editing is the capability Google built the launch around.

The audio trap in the Artificial Analysis numbers

Artificial Analysis publishes each video board twice — with audio and without. Almost nobody quoting these numbers says which one they mean, and the two disagree sharply.

Here is Gemini Omni Flash on both views, read 28 August 2026:

The image-to-video row is the one to sit with. First place becomes fourth depending on which view you open, and a post that says "number one on image-to-video" without naming the view is not reporting a result, it is picking one.

Both views are legitimate. Audio is a real part of what these models now produce, and a model that generates a strong picture with weak sound genuinely should rank lower on a board that counts sound. But the gap between first and fourth is exactly the kind of detail that disappears in a headline.

The editing board, dated properly

There is a story circulating that Gemini Omni Flash debuted at number one on video editing and was later pushed to second by MiniMax H3. That was true. It stopped being true a while ago.

• Late July 2026 — already 2nd, at 1,122 behind MiniMax H3 at 1,130
• 6 August 2026 — 2nd, 1,123 against 1,132
• 28 August 2026 — 3rd, at 1,124, with Wan 3.0 first at 1,189 and MiniMax H3 second at 1,128

Alibaba's Wan 3.0 now leads two of the three Artificial Analysis boards, which is the least-reported fact in this whole subject. On arena.ai's video-edit board Gemini Omni Flash sits fourth, behind two Seedance variants and MiniMax H3 — and its score there actually rose over the period, from 1,347 to 1,358. It did not get worse. It got passed.

That is the honest shape of this market. Four labs are trading places, the leader depends on which board and which view you open, and nobody holds all of it.

What Google itself claimed

Nothing. Not a single Elo, benchmark or leaderboard placing appears in Google's announcement, in the model card, or in the API documentation. The launch is a feature list and a price.

We think that is worth reading as a signal rather than an omission. A lab that had a decisive benchmark result usually leads with it. Google led with scene extension, frame control and a cheaper draft tier — the things you would emphasise if your model's advantage were in workflow rather than in a head-to-head win. The one number Google did publish is a price, and we covered what the launch actually changed separately.

There is also, as of today, no independent head-to-head comparison of 1.1 against its predecessor. Every article you can find is a rewrite of the announcement. Anyone telling you how much better 1.1 is has not measured it, because there is nothing yet to measure it with.

So what should you actually run today

Here is the part the boards are quietly telling you: the model topping them is not the new one. It is the predecessor. Until 28 August 2026 that was a real constraint, because 1.1 was not yet callable here. It is now, which changes the question from what you can run to what you should.

Omni Flash, the model behind every without-audio first place quoted above, still runs on our API. But Gemini Omni 1.1 Flash has been callable here since 28 August 2026 at $0.09 a second at 720p, and the argument for it was never going to be a ranking. It is the things the older model cannot do at all: image-to-video and start-end-frame-to-video, both new to this family on E2X, reference-to-video taking up to seven reference images, a 360p draft tier at $0.0297 a second, and clips that run to ten seconds. See the launch write-up for what actually changed between them.

Now the uncomfortable arithmetic. Veo 3.1 Fast costs a cent a second, one ninth of what Omni 1.1 Flash does, and it is plainly the weaker model, sitting around ninth once audio is counted.

But look at the ratio rather than the ranking. Nine times the money against a few dozen Elo points, on a scale where the whole top ten spans about 120. One eight-second shot is the difference between eight cents and seventy-two, and the sixty-four cents in between buys eight more attempts at the cheap tier. Most shots do not need the best model. They need three more tries.

The premium stops being optional when a job needs continuity: a character who has to survive between shots, a specified opening and closing frame, a clip that gets extended later. There a weaker model does not degrade gracefully, it simply fails. Veo 3.1 Fast will still handle plenty of that — image-to-video, reference-to-video, first and last frame — so let a real failure, not a leaderboard, be the thing that moves you up.

Prices and product terms can change. Check each provider's current pricing before making a purchasing decision. This is a scoped comparison, not a claim that E2X is the cheapest option in every configuration.

How to read a video leaderboard without getting burned

Four habits, and they will serve you long after these particular numbers are stale:

1. Check the model string, not the marketing name. "Gemini Omni Flash" and "Gemini Omni 1.1 Flash" are different products, and almost every article published this week uses the first name for the second model.
2. Ask which view. With audio or without changes first place into fourth on one of the boards above.
3. Read the vote count and the interval. A three-point lead on 1,762 votes is a coin toss dressed as a result.
4. Check the date on the standing, not the date on the article. A piece published yesterday can be quoting a July snapshot, and two of the most-shared claims in this subject are exactly that.

Pick by job from text-to-video or reference-to-video, or read the machine-readable spec — Omni 1.1 Flash publishes one with every parameter and the current price. The whole lineup is in our catalog.

Frequently asked questions

Is Gemini Omni 1.1 Flash number one on the video leaderboards?

On one board, narrowly and provisionally. It sits first on arena.ai's Text-to-Video Arena at 1515 ±16, but only three Elo points ahead of its own predecessor on 1,762 votes against 19,830, with overlapping confidence intervals — a statistical tie. It is second on arena.ai Image-to-Video and does not appear on Artificial Analysis's boards at all.

Does Artificial Analysis rank Gemini Omni 1.1 Flash?

No. As of 28 August 2026 Gemini Omni 1.1 Flash appears on none of Artificial Analysis's three video leaderboards. Every Artificial Analysis figure circulating for "Omni Flash" refers to the earlier Gemini Omni Flash preview model, which shipped in June 2026.

Why do different sites give different rankings for the same model?

Mostly because Artificial Analysis publishes each board twice, with audio and without, and most articles do not say which they used. Gemini Omni Flash is first on image-to-video without audio and fourth with it. Board choice, date and the with-audio question explain nearly every contradiction you will find.

Is Gemini Omni Flash still number one on video editing?

No. It debuted at the top, was second behind MiniMax H3 through late July and early August, and sits third as of 28 August 2026 with Wan 3.0 first. Its score did not fall — it rose slightly, from 1,347 to 1,358 on arena.ai — but other models improved faster.

Did Google publish any benchmark results for Gemini Omni 1.1 Flash?

No. Google's announcement blog, the API documentation and the DeepMind model card contain no Elo, benchmark or leaderboard claim for the model. The launch material covers features, availability and pricing only.

Which video model is the best right now?

There is no single answer, which is the honest reading of the boards. Alibaba's Wan 3.0 leads two of the three Artificial Analysis boards, MiniMax H3 leads arena.ai's image-to-video, ByteDance's Seedance 2.5 leads its video-edit board, and Gemini Omni Flash leads the two Artificial Analysis boards in their without-audio view. Pick by the job you have, not by a headline.

Is Gemini Omni 1.1 Flash worth the price over a cheaper model?

For single clips with no editing or continuity requirement, usually not. Veo 3.1 Fast runs at $0.01 per second against Gemini Omni 1.1 Flash at $0.09 on our API as of 28 August 2026 — a nine-times gap for a difference of a few dozen Elo points. The premium earns its keep on reference consistency, first-and-last frame control and scene extension, where a weaker model does not just look worse but fails the task.