A codename, paper-volcano, appeared in Google Cloud’s AI quota list on 17 November (NZ time). Posters soon began calling it Nano Banana Pro 2, and on 21 November an X post by Astro Polo (@astropol0) said a launch could come sooner than expected. Google hasn’t said what paper-volcano is. The quota entry is the one claim in the chain that several people reported independently.
What the trail shows
The earliest record I found is from an automated tracker called AI Tracker Bot. According to AREX’s write-up of the posts, it flagged at 12:49pm on 17 November that two unannounced models had appeared in Google Cloud’s quota list: gemini-embedding-2-flash, and paper-volcano, a codename with a quota for generating content. The bot added that the developer hadn’t confirmed either.
About two hours later, an X user called lyra named paper-volcano as Nano Banana Pro 2 and wrote that the checkpoint didn’t perform very well in their tests. AREX describes the attached screenshot as a panel with a base_model:paper-volcano row. No prompts, sample images or scores came with the post.
Other accounts added to it the same day. JEX_kcaj posted from their own Google Cloud quota page that paper-volcano is capped at 20 requests a minute and a second codename, spicy-mayo, at 250. A card from StatsWire, relayed by xiax0603, listed paper-volcano among four new quota dimensions on aiplatform.googleapis.com, next to gemini-embedding-2-flash, gemini-4-hydrogen-qcd and veo-experimental. The card said the quota dimensions were confirmed on Google Cloud and the Nano Banana Pro 2 label wasn’t.
17 Nov
First quota record (NZ time)
20 a minute
Limit one account reported
12 days
Nano Banana 2.1, first Flow trace to release
0
Google statements naming paper-volcano
Astro Polo’s summary came on 21 November. It says paper-volcano has appeared in Google Flow and Google Cloud, that people spotted it in production code and quotas, and that the update will focus on editing, consistency and image quality. It passes on an early tester’s view that the checkpoint wasn’t strong and calls that unverified. The text names no source for any of it.
What a quota entry proves
A base_model:paper-volcano quota dimension shows that Google Cloud has set aside call capacity for something with that name. It doesn’t show who built it, what it produces or how well it works. The batch it appeared in also holds an embedding model and veo-experimental, so a codename in that list could belong to any of Google’s model families. According to AREX, xiax0603 called it a newly exposed base-model codename, not an image model.
The name isn’t Google’s
Two word orders are circulating. Lyra and JEX_kcaj wrote Nano Banana Pro 2, and Astro Polo wrote Nano Banana 2 Pro. Google’s image models are named Nano Banana Pro, Nano Banana 2, Nano Banana 2 Lite and Nano Banana 2.1. Neither word order is one of them.

Nano Banana Pro is Gemini 3 Pro Image. Its Gemini API page shows a latest update of November 2025, and Google Cloud lists it as generally available since 28 May 2026. Nano Banana 2.1 arrived on 6 October as the newer, faster model, based on Gemini 3.6 Flash.
Why a Pro successor is a fair guess
Google’s own model card for Nano Banana 2.1 gives scores for the three models. Nano Banana Pro scores lower than 2.1 on every row, and the areas Astro Polo says the new model targets, editing and consistency, are among them.
| Measure | Nano Banana 2.1 (thinking) | Nano Banana 2 | Nano Banana Pro |
|---|---|---|---|
| Overall preference | 1050 | 990 | 935 |
| Infographic design | 1048 | 961 | 912 |
| Infographic factuality | 0.521 | 0.179 | 0.265 |
| General editing | 1026 | 938 | 939 |
| Multi-character consistency | 1106 | 978 | 1011 |
Both Nano Banana Pro and 2.1 sit on Gemini 3 bases. Gemini 4 Carbon, Hydrogen and Neon went to general availability on 10, 12 and 18 November, and Google’s Gemini API release notes, current to 19 November, list no image model after Nano Banana 2.1 on 6 October. That gap is why the rumour is plausible. It isn’t evidence about paper-volcano.
Last time: 12 days from a Flow string to a release
Nano Banana 2.1 left a documented trail. On 24 September, TestingCatalog posted a screenshot of Nano Banana 2.5 Flash traces in Google Flow. By 27 September the same Flow build labelled the model 2.1. On 5 October users posted screenshots of 2.1 in Flow’s image picker, and TestingCatalog noted the output looked like Pro’s, so it couldn’t tell whether requests reached a new model. On 6 October Google’s release notes announced it as generally available and DeepMind published a model card. The rumoured number, 2.5, didn’t survive.

The Arena codename spicy-mayo, which testers tied to Nano Banana 2.5 from September, has never been matched to a Google model. JEX_kcaj’s post shows it in the quota list on 17 November, six weeks after 2.1 shipped. paper-volcano has a quota entry and nothing else on that list. I found no screenshot of it in Flow, no Google documentation and no model card.
Claims with no source behind them
Four claims in the chain rest on posts that show nothing checkable.
- In Flow: pankajkumar_dev wrote on 20 November that it was spotted in Google Flow under the name Paper Volcano, and a Threads post says the same. Neither names a build. Google’s Flow page lists Nano Banana Pro and Nano Banana 2.1 as its image models.
- In Arena: the same posts say it’s being tested there. I found no Arena entry for paper-volcano.
- Next week: pankajkumar_dev says the model is in its final stages and could be seen next week. The post names no source.
- Poor results: lyra’s remark is a single tester’s view with no prompts or samples, and Astro Polo repeats it as an early tester’s claim.

What would settle it
Google releases leave a paper trail anyone can check. Any one of these would count:
- A new model ID and date in the Gemini API release notes.
- A model page on Google Cloud with a launch stage and release date.
- A DeepMind model card.
- A pricing row.
- A picker entry in Flow or the Gemini app that Google documents.
Until one of those appears, paper-volcano is a codename with a quota. Nothing changes for anyone building on Nano Banana today. Google’s docs call gemini-nano-banana-2.1 the model to use for new projects, and gemini-3-pro-image stays available.
What Pro means in Google’s naming
When Nano Banana 2 launched in February, Google described Nano Banana Pro as the choice for high-fidelity tasks that need maximum factual accuracy, and Nano Banana 2 as the choice for rapid generation, precise instruction following and image-search grounding.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.




