Google Releases Nano Banana 2.1: Steadier AI Image Editing at Half the Developer Price
Google released Nano Banana 2.1 on October 6, a new image generation and editing model that replaces version 2.0 from February. The company says the update improves visual design, mask-based editing, and subject consistency across the board, producing more natural-looking images.
Subject consistency is the headline upgrade: the model can process up to 14 reference images at once while tracking as many as four characters and ten objects in a single frame. Faces and objects stay recognizable across multiple rounds of edits instead of drifting into something unrecognizable after two or three passes — a fix that matters for anyone doing iterative image work. Built on Gemini 3.6 Flash, the model supports 1K, 2K, and 4K output, stretches aspect ratios as wide as 8:1, fixes tiling artifacts on wide panoramas at high resolution, and renders text and infographic layouts more accurately.
Nano Banana 2.1 keeps the search-grounding capability introduced in version 2: it can consult Google Web Search and Image Search to verify real-world details before rendering, so images depicting real objects and events come out more accurate. Not every AI image tool does that, and Google is treating it as a meaningful differentiator. Developers also get adjustable inference "thinking" levels — minimal, medium, and high — to trade speed against quality per task.
On text-to-image benchmarks, version 2.1 scored 1,050 ELO points, up from 990 for its predecessor and 935 for the one before. Google is candid that it is still not the company's best: Nano Banana Pro usually produces noticeably better images in practice, but that is the flagship tier — the company describes 2.1 as Pro's "more efficient counterpart."
For developers, the price cut may be bigger news than the pixels. Standard paid-tier API rates for image tokens dropped from $60 to $30 per million. A 1K image now costs 3.36 cents, down from 6.70 cents; a 4K image falls from 15.10 to 7.56 cents. Batch processing jobs get another 50 percent off across the board, bringing 1K generations down to 1.68 cents per image.
The rollout is sequential: the Gemini app, Search's AI Mode, Google Ads, and developer platforms including AI Studio, Flow, Stitch, and the Gemini Enterprise Agent Platform. Robby Stein, Google's VP of Product for Search, announced on X that the model is already live in AI Mode — users tap the banana icon under the Search box in the Google app to start creating. The model also carries C2PA Content Credentials for digital origin tracking.