Nano Banana 2.1 in Google Flow: What's Confirmed, What Isn't

Nano Banana 2.1 is showing up as a selectable model inside Flow, and as of October 6, 2026 that is the entire public record. The company has published no announcement, no model card and no API identifier for it. The name sits in a picker next to Nano Banana Pro and Nano Banana 2 Lite, and it is a 0.1 bump on Nano Banana 2 — the model shipped as Gemini 3.1 Flash Image. There is a real story here, but it is a story about a deployment sighting, not a launch, and the difference matters if you are deciding what to build on this quarter.

Two things are simultaneously true. A new model label is genuinely reachable in a Google product, which is more than a rumor. And nothing about that label has been documented by the company that made it, which is less than a release. Everything below separates those two.

What actually changed, in order

The trail starts in the Flow web build. On September 24, 2026, references inside it pointed to a model called Nano Banana 2.5 Flash — a name that implied a larger jump, closer to a half-generation. On September 27, TestingCatalog reported that the same build had relabeled that reference to Nano Banana 2.1. The naming change was the first signal, and it pointed downward in ambition: a 2.1 is an iterative release, not a generational one.

Then on October 5 the label stopped being an internal string and became a menu item. Users posted screenshots of the Flow image-generation interface listing Nano Banana Pro, Nano Banana 2 Lite and Nano Banana 2.1, with 2.1 selected. One of those posts associated the model with the internal name "beluga" — a single-source claim, reproduced by aggregators but not corroborated by Google or by a second independent reporter. Treat the codename as reported, not established.

The same day, TestingCatalog added the caveat that makes this story worth writing carefully: the outputs it saw from Nano Banana 2.1 looked like what Nano Banana Pro produces, so it could not establish that requests were routed to a genuinely new model rather than to an existing one under a new label. Google has not clarified either way.

What is confirmed, and what is not

Confirmed: the string "Nano Banana 2.1" appears as a selected model in Google Flow's image-generation UI in screenshots published October 5, 2026. That is it. That is the whole confirmed set.

Not confirmed, and each of these is a thing a reader will reasonably assume:

• No Google announcement. There is no blog post, no Workspace update, no DeepMind model card, and no entry in the Gemini API changelog, which was last updated October 1, 2026 and lists only Nano Banana 2 (Gemini 3.1 Flash Image), Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image) and Nano Banana Pro (Gemini 3 Pro Image).

• No API identity. Nano Banana 2.1 has no published model identifier. Google's developer surface exposes gemini-3.1-flash-image, gemini-3.1-flash-lite-image and gemini-3-pro-image. There is nothing for a developer to call.

• No price. No per-image rate, no token rate, no statement about which Google AI plan tier includes it. Flow's own interface showed one user that generating with it would consume zero AI credits, which tells us about that account's plan, not about pricing.

• No benchmark. No entry on Artificial Analysis, no LMArena board, no vendor eval. Any number attached to Nano Banana 2.1 today would be invented.

• No scope. The screenshots do not establish whether the label is available to all Flow users, to a test group, or to the specific accounts that posted. A limited rollout and a broad one look identical from a single screenshot.

• No verified relationship to Nano Banana Pro. The "it looks like Pro" observation is a visual impression from a handful of outputs, not a routing trace.

What makes the unconfirmed list unusually long is that it is not a case of a model known to be coming with details still under embargo. The label is inside a shipping Google product, which means a build exists in front of users — yet every artifact that normally accompanies even a quiet release is absent.

The name it nearly had, and why it was walked back

The abandoned "Nano Banana 2.5 Flash" label is the most informative detail in the whole sequence. We covered that string when it first surfaced, in Nano Banana 2.5 Flash leak: Google's next image model is in the works, and at the time the reasonable read was that Google was preparing a half-step up the tier ladder. The September 27 relabel to 2.1 says the opposite. Whatever ships under this name is positioned as a point release against Nano Banana 2, not as a new tier between Nano Banana 2 and Nano Banana Pro.

That reframing matters commercially. A "2.5 Flash" would have been a new price-performance point worth waiting for. A "2.1" is much more likely to be a quality and consistency refresh on the same tier — the kind of change that improves outputs without moving the cost line. If you were holding a procurement decision open for a cheaper Flash-class model from Google, the renaming is evidence you should stop holding it.

What a 2.1 would have to beat

The comparison that will matter when this lands is against Google's own line, and Google's line is unusually well measured — which is exactly why the absence of a benchmark for 2.1 is conspicuous rather than merely early.

On Artificial Analysis, as of October 6, 2026, Nano Banana 2 (Gemini 3.1 Flash Image) sits at Elo 1,125 on the text-to-image board — sixth of the ranked entries, behind the two GPT-Image-2.5 variants, GPT-Image-2 (high), Grok Imagine Image 2.0 and MAI-Image-2.6, and ahead of Google's other two image models. On the image-editing board the same model sits at Elo 1,108, which is now eighth — the editing board has been reshuffled since the September GPT-Image-2.5 entries landed. Nano Banana Pro is at Elo 1,102 on text-to-image and 1,099 on editing. Nano Banana 2 Lite is at Elo 1,096 on text-to-image. Artificial Analysis lists Nano Banana 2 at $67 per 1,000 images, Nano Banana Pro at $134, and Nano Banana 2 Lite at $33.60 — those are the operator's own measured price points, not vendor quotes.

Read together, that board says something specific about where a 2.1 would have to land. Nano Banana 2 already trades quality for price at the volume end of Google's own family, and Nano Banana Pro already owns the top of it. A 2.1 that only matches Pro visually, as the early Flow outputs reportedly do, would collapse three tiers into one. A 2.1 that beats Nano Banana 2's 1,125 while holding its per-image price is the version worth waiting for. We cannot yet tell which one Google built.

The one thing you can do about an unannounced model

There is nothing to integrate yet, so the useful move is preparation rather than adoption. The generic problem the Nano Banana family creates is that Google exposes it across surfaces with different lifecycles: the Gemini app leads, Flow follows, and the developer API trails with its own identifier and its own GA date. Nano Banana 2 followed exactly that pattern — a February 26, 2026 consumer and Workspace rollout, and a developer GA on May 28 under the id gemini-3.1-flash-image. If 2.1 repeats it, the picker you can see today is months ahead of the endpoint you would actually call.

When it does reach an API, the practical requirement for most teams is that adding it not mean a second contract and a second billing relationship. That is the case OrcaRouter is built for: one endpoint across 200-plus models, with each provider's list price passed through at 0% markup, so a Google price change is live on our side the same day rather than on a renewal cycle. Our catalogue already carries Google's image line at those rates — including google/gemini-3.1-flash-image-preview and google/gemini-2.5-flash-image — and the Google image generations that ship as stable API models get added the same way. To be explicit about the state of things today: Nano Banana 2.1 is not on OrcaRouter, because it is not on anything except a Flow model picker.

What to watch, and what would change this article

Three events would move Nano Banana 2.1 from sighting to release, and they will not arrive together. The first is an entry in the Gemini API changelog or a new model card on ai.google.dev, because that is what gives it an identifier and a price. The second is an Artificial Analysis entry, because that is what turns a visual impression into a number. The third is a Google blog post, because that is what establishes scope — who gets it, on which plan, in which countries.

Until at least one of those lands, the honest summary is the one at the top of this page. There is a new label in a Google product. It is real enough to screenshot and empty enough to say nothing about. If you need image generation from Google today, Nano Banana 2 remains the measured, documented, callable choice — and the sensible thing to do with Nano Banana 2.1 is watch it, not plan around it.