Half the Price, More Banana: Google's Nano Banana 2.1 Rewrites the Image-Gen Value Chart

A week after retiring the original Nano Banana API, Google has shipped its successor — and the pricing story is the headline. Nano Banana 2.1, released October 6 and built on the Gemini 3.6 Flash foundation, generates and edits images at roughly half the cost of its predecessor, while posting benchmark scores that in several editing categories beat Nano Banana Pro, Google’s flagship image model at nearly four times the price.

For anyone building image generation into a product, this is the rare update where the press release undersells the spreadsheet.

What shipped

Nano Banana 2.1 replaces Nano Banana 2 (released February 2026) as Google’s mainstream image generation and editing model. The lineage matters here: the “Flash-tier” Banana models have always been the workhorses — fast, cheap, good enough for most jobs — while the Pro models chase maximum fidelity at premium prices. Version 2.1 keeps that positioning but pulls the quality curve up hard enough that the gap to Pro has become genuinely hard to justify for many workloads.

Google describes 2.1 as “the more efficient counterpart” to Pro, and claims improvements “across the board” over previous versions in visual quality, text rendering, character consistency across conversation turns, wide panoramas, and infographics.

The price cut, in numbers

The per-image pricing is where 2.1 makes its case:

Model	1K image	2K image	4K image
Nano Banana 2.1	$0.0336	$0.0504	$0.0756
Nano Banana 2 / Gemini 3.1 Flash Image	$0.0670	$0.1010	$0.1510
Nano Banana 2 Lite	$0.0336	—	—
Nano Banana Pro / Gemini 3 Pro Image	$0.1340	$0.1340	$0.2400

A 1K image drops from 6.70 cents to 3.36 cents — a 49.9% cut. A 4K image falls from 15.10 to 7.56 cents. And against Nano Banana Pro, the comparison is stark: Pro charges 13.40 cents per 1K image, exactly four times what 2.1 costs.

That Lite-tier price point is significant. Nano Banana 2 Lite was Google’s bargain option for 1K images only; 2.1 now matches Lite’s price while offering 1K, 2K, and 4K output, Flash-tier speed, and the full feature set. The bargain tier and the mainstream tier have effectively merged.

Benchmarks: efficiency champion, with an asterisk

Google’s technical report shows 2.1 (with thinking enabled) posting strong numbers across text-to-image and editing evaluations:

• Text-to-Image Overall Preference: 1050 ± 14 — ahead of Gemini 3.1 Flash Image (990) and Gemini 3 Pro Image (935)
• Infographic Design: 1048 ± 17, with an Infographic Accuracy score of 0.521 — nearly triple Gemini 3.1 Flash Image’s 0.179
• Multi-Character Consistency: 1106 ± 14 — the standout number, beating Pro (1011) by a wide margin
• Mask/Ink-Based Editing: 1049 ± 15 vs Pro’s 927
• Stylization and Multi-Reference Editing: 1062 and 1066 — both category-leading

The Decoder’s hands-on testing adds a dose of realism, though: Nano Banana 2 also matched Pro in those same tests, and in practice Pro “often still produces noticeably better images.” In The Decoder’s own benchmark prompt, 2.1 followed the complex compositional instruction closely but struggled with scale, rendering a horse “more like a pony.” Benchmark elo and perceptual quality are correlated, not identical.

Capabilities beyond the scorecard

Several features matter more for production use than raw preference scores:

• 14 reference images per request. The model can ingest up to fourteen reference images simultaneously and keep up to four characters and ten objects visually consistent across them. For storyboarding, product catalogs, and brand work, this is the difference between “image generator” and “asset pipeline.”
• Google Search grounding. The model connects to Google Search, letting it pull factual grounding for infographics and text-heavy images — a meaningful advantage for the accuracy-sensitive infographic use case where it scores 0.521 vs 0.179 for the previous Flash model.
• Thinking levels. Minimal, medium, and high thinking modes let developers trade latency and cost against quality — a knob most competitors don’t expose.
• Multi-turn consistency. Character consistency is maintained across conversation turns, not just within a single generation.

Rollout and the October 29 deadline

The model runs at Flash-tier speed and is already rolling out across Google’s surface areas: the Gemini app, AI Mode in Google Search, Google AI Studio, Flow by Google, Stitch by Google, Google Ads, and the Gemini Enterprise Platform.

Developers should note the transition timeline: Google will shut down the previous model, gemini-3.1-flash-image (Nano Banana 2’s API identifier), on October 29, 2026. That gives anyone still on the old API three weeks to migrate — and since 2.1 is both better and half the price, the migration is one of the rare forced upgrades that’s also an unambiguous win.

Analysis: the commoditization squeeze

The strategic read here is straightforward: Google is using price as a weapon in the image generation market. The frontier quality fight — where Nano Banana Pro, OpenAI’s GPT-image models, and the coming Gemini 4 Argon compete — gets the headlines. But the volume business, the one that shows up in production APIs and enterprise pipelines, is won on price-performance. Matching Lite pricing at full Flash capability is a deliberate squeeze on every mid-tier competitor.

It also telescopes Google’s roadmap. Nano Banana Pro still runs on Gemini 3 Pro, and 2.1’s benchmarks already nip at it in editing tasks. When the announced Gemini 4 Argon arrives with a Pro-tier image model on top, the quality gap at the high end should reopen — at which point the Flash tier will have absorbed most of what used to be Pro-level quality, at a quarter of the price.

For developers, the practical takeaway: if you’re paying Pro prices for editing, consistency, and infographic work, test 2.1 against your own prompts this week. The benchmarks say you can probably switch. Your eyes will tell you whether you should.