The price cut is real, but the comparison needs a footnote

Nano Banana 2.1 halves Google's image-output token rate from $60 to $30 per million tokens. That takes a 1K image from the previous model's $0.067 to $0.0336, and a 2K image from $0.101 to $0.0504.

The 4K row is where a simple launch headline can outrun the billing table. Google's current Gemini API page lists a standard 4K output at $0.113 and a Batch 4K output at $0.0567. So the accurate comparison is not standard to standard: the new model's asynchronous Batch price for 4K is now below the old model's standard price for 1K, which was $0.067.

That distinction matters. Batch work is queued rather than returned as an interactive response. It suits a catalogue refresh, a backlog of marketing variants or any production run that can wait. It is not a like-for-like replacement for a person editing an image in real time.

Resolution is no longer the expensive decision

For high-volume builders, the important change is the slope of the cost curve. Under the previous Nano Banana 2 pricing, moving from 1K to 4K raised image-output cost from $0.067 to $0.151. With 2.1, the current standard figures are $0.0336 at 1K, $0.0504 at 2K and $0.113 at 4K. Batch cuts each of those in half.

At 10,000 outputs, that makes the image portion roughly $336 at 1K, $504 at 2K or $1,130 at 4K on standard pricing. The same run through Batch is about $168, $252 or $567. A team that used to render low-resolution drafts and selectively upscale finals can now consider generating more final-size assets directly, especially when turnaround is not urgent.

But resolution is not quality. Paying for more output pixels does not guarantee a cleaner composition, accurate typography or a usable edit. The cheaper rate lowers the penalty for asking for 4K; it does not remove the cost of failed generations.

The request costs more than the image row

Google bills image output separately from input and reasoning. Nano Banana 2.1 charges $1.50 per million input tokens and $7.50 per million text-and-thinking output tokens on the standard tier. Search grounding can add charges after a shared monthly allowance. There is no free API tier for this model.

That means $0.0336 or $0.113 is not necessarily the total invoice for a call. Prompts, uploaded references, model thinking, grounded searches and retries all add to the real cost. A workflow that uses 14 reference images and several editing turns may spend more on context and failed attempts than the clean per-image table suggests.

The right unit for a production team is cost per accepted asset, not cost per generated file. If two out of five generations are usable, the effective price is already 2.5 times the output line item before labour enters the calculation.

2.1 is an operational update, not just a discount

Google describes Nano Banana 2.1 as the efficient counterpart to Nano Banana Pro. It supports 1K, 2K and 4K output, configurable thinking levels, grounding with Google Web and Image Search, and up to 14 reference images. The model documentation says it can preserve up to four characters and ten objects across a multi-image workflow.

The release also targets the unglamorous failures that make image models expensive in practice: inconsistent subjects over repeated edits, weak text layout, and tiling artifacts in very wide panoramas. Google says 2.1 improves all three and supports aspect ratios as wide as 8:1.

Those claims are useful because they point to measurable work. They are not independent evidence. A benchmark win on Google's own prompt set says less than the number of retries a designer needs to keep a product, character or wordmark stable across twenty assets.

The real competitive move is price discipline

Google is segmenting its image stack more clearly. Nano Banana 2.1 is the Flash-priced default for capable generation and editing. Nano Banana 2 Lite stays at 1K for low-latency volume work. Nano Banana Pro remains the expensive option, with standard image output priced at $0.134 for 1K or 2K and $0.24 for 4K.

That creates a practical ladder rather than one model pretending to cover every job. It also puts pressure on image API vendors whose pitch depends on a fixed per-generation price. When the cost of a high-resolution output falls toward the old low-resolution tier, orchestration, editing controls and acceptance rate become more important than access to a generator.

The headline is compelling because the economics have moved. The useful reading is narrower: Google's cheapest 4K route now costs less than yesterday's ordinary 1K route, provided the job can run in Batch. For interactive work, the new 4K price is still materially higher than the old 1K price. The cut is substantial without pretending the service levels are identical.

Sources

  1. [1] Google AI for Developers — Gemini API pricing (checked Oct 7, 2026)Read source
  2. [2] Google AI for Developers — Gemini Nano Banana 2.1 model documentationRead source
  3. [3] Google DeepMind — Nano Banana 2.1 model card (Oct 6, 2026)Read source
  4. [4] The Decoder — “Google's new image model Nano Banana 2.1 generates better images for less money”Read source