Google releases Nano Banana 2.1 image AI, API image output at about half the price
In short
Google released Nano Banana 2.1, an image generation and editing AI model built on Gemini 3.6 Flash, on October 6 US time and is rolling it out across the Gemini app, Google AI Studio, the Gemini API and Google Search's AI mode, among other products. The company's human evaluations put it above the earlier Nano Banana 2 and the higher-tier Nano Banana Pro, while paid-tier image output costs about half as much per image as Nano Banana 2. Google lists small text rendering, character consistency and spatial left-right confusion as remaining limitations.
Read the full story 3 min read
Google released the image generation and editing AI model Nano Banana 2.1 on October 6 local time and began rolling it out the same day, according to the three technology outlets reporting the launch. The model is being offered in the Gemini app, Google AI Studio, the Gemini API, Google Search's AI mode, Google Ads, Google Flow and Google Stitch, and ITmedia AI+ adds the Gemini Enterprise Platform to that list. [ 1 , 2 , 3 ]
Nano Banana 2.1 is the successor to Nano Banana 2 and is based on Gemini 3.6 Flash. The outlets describe it as a high-efficiency model that keeps Flash-level speed and cost efficiency, and as a more efficiency-focused counterpart to the higher-tier Nano Banana Pro (Gemini 3 Pro Image). [ 1 , 2 , 3 ]
In Google's human comparison evaluations cited by the outlets, Nano Banana 2.1 scored 1,050±14 with Thinking enabled on overall preference for text-to-image generation, above Nano Banana 2's 990±7 and Nano Banana Pro's 935±8. On multi-character consistency it scored 1,106, against 978 for Nano Banana 2 and 1,011 for Nano Banana Pro. ITmedia AI+ also reports an infographic factual accuracy score of 0.521, against 0.179 and 0.265, and says the new model beat both earlier models on all seven evaluated items, including mask and ink-based editing at 1,049 versus 965 and 927. [ 1 , 2 , 3 ]
On pricing, ITmedia AI+ reports that the paid Gemini API tier charges $30 per million tokens for image output, with no free tier, plus $1.50 per million input tokens and $7.50 for text and thinking output. Per image, that comes to $0.0336 at 1K, $0.0504 at 2K and $0.0756 at 4K, roughly half Nano Banana 2's $0.067, $0.101 and $0.151. The same outlet notes the input price rose from $0.50 to $1.50 per million tokens, and that Batch API rates are half. PC Watch says it is listing API prices but the figures are not present in the text we could read. [ 1 , 2 ]
Feature updates listed by the outlets include higher image quality and realism at 1K, 2K and 4K, fixes to tiling artifacts at wide and panoramic aspect ratios, improved text rendering and infographic layouts, support for up to 14 reference images while keeping up to four characters and ten objects consistent, grounding through Google web and image search, and selectable Thinking levels (minimal, medium and high, with medium as the default according to PC Watch). PC Watch also reports that the Nano Banana 2 model ID “gemini-3.1-flash-image” is deprecated and its service ends on October 29. [ 1 , 3 ]
As known limitations, Google lists blurred rendering of small text and long paragraphs, character consistency that is not always perfect between input and output images, occasional confusion over spatial positions such as left and right, and remaining limits in higher-level abilities such as world knowledge, 3D reasoning and factual recognition. GIGAZINE adds that the knowledge cutoff of the underlying Gemini 3.6 Flash is March 2026, with knowledge in some domains available only to January 2025. [ 1 , 2 , 3 ]
The documents disagree on one point of background: PC Watch says Nano Banana 2 appeared in February 2026, while GIGAZINE says it was officially released in May 2026. All three accounts are based on Google's announcement rather than independent testing. [ 1 , 3 ]
Why it matters
The release puts a cheaper, efficiency-focused image model above Google's own higher-tier model in the company's benchmark scores, which the tech outlets present as a shift in how lower-cost models compare with premium ones. A roughly halved image output price on the paid API tier lowers the cost for developers and businesses embedding image generation, though the input price rose. The coverage is limited to Japanese tech media reporting Google's announcement, so the practical effect is not yet independently verified.
Key facts
- Google released the image generation and editing AI model Nano Banana 2.1, a successor to Nano Banana 2, and said it is being rolled out across the Gemini app, Google AI Studio, the Gemini API, Google Search's AI mode, Google Ads, Google Flow and Google Stitch. [ 1 , 2 ]
- Nano Banana 2.1 is based on Gemini 3.6 Flash and is positioned as a more efficient counterpart to the higher model Nano Banana Pro (Gemini 3 Pro Image). [ 1 , 2 , 3 ]
- In Google's human comparison evaluations, Nano Banana 2.1 scored 1,050 with Thinking enabled, above Nano Banana 2's 990 and Nano Banana Pro's 935. [ 1 , 2 ]
- On multi-character consistency, Nano Banana 2.1 scored 1,106, against 978 for Nano Banana 2 and 1,011 for Nano Banana Pro. [ 1 , 2 , 3 ]
- Per ITmedia AI+, paid-tier Gemini API pricing is $30 per million tokens for image output, or $0.0336 per 1K image, $0.0504 per 2K image and $0.0756 per 4K image, roughly half Nano Banana 2's rates, while the input price rose from $0.50 to $1.50 per million tokens. [ 2 ]
- Nano Banana 2.1 accepts up to 14 reference images and can keep up to four characters and ten objects consistent. [ 1 , 3 ]
- Google lists blurred rendering of small text and long paragraphs, imperfect character consistency between input and output images, occasional confusion of spatial positions such as left and right, and limited world knowledge and 3D reasoning as known limitations. [ 1 , 2 , 3 ]
- PC Watch reported that the Nano Banana 2 model ID “gemini-3.1-flash-image” is deprecated and its service ends on October 29. [ 1 ]
Confirmed by several sources
- Google released Nano Banana 2.1, an update to its Nano Banana 2 image generation and editing model. [ 1 , 2 , 3 ]
- The new model is based on Gemini 3.6 Flash. [ 1 , 2 , 3 ]
- The model is provided through the Gemini app, Google AI Studio, the Gemini API and Google Search's AI mode, among other products. [ 1 , 2 ]
- In Google's human evaluations, Nano Banana 2.1 scored 1,050 with Thinking enabled, above Nano Banana 2 (990) and Nano Banana Pro (935). [ 1 , 2 ]
- Nano Banana 2.1 scored 1,106 on multi-character consistency, versus 978 for Nano Banana 2 and 1,011 for Nano Banana Pro. [ 1 , 2 , 3 ]
- Google's listed known limitations include poor rendering of small text and long paragraphs, imperfect character consistency, occasional spatial left-right confusion and limited higher-level abilities. [ 1 , 2 , 3 ]
Still unclear
- When Nano Banana 2 was released. PC Watch says Nano Banana 2 appeared in February 2026, while GIGAZINE says it was officially released in May 2026; the documents disagree.
- The full Gemini API price table for Nano Banana 2.1 referred to by PC Watch. PC Watch states prices are listed below, but the figures are not present in the text we could read; pricing details come from ITmedia AI+ only.
- The “minimal/medium/high” Thinking levels and the default of medium. Stated only by PC Watch and not corroborated by the other documents.
- Google's announcement itself, including any statement from the company beyond what the outlets report. All three documents are Japanese tech outlets reporting the release; no official document is among the sources.
What local media are saying
Timeline, local time
- Google releases Nano Banana 2.1 and begins rolling it out on its products, according to the reports. [ 1 , 2 ]
- PC Watch publishes its report on the release, describing Nano Banana 2.1 as a cheaper model that outperforms the higher-tier model. [ 1 ]
- ITmedia AI+ publishes its report, detailing Gemini API pricing and per-image costs. [ 2 ]
- GIGAZINE publishes its report, listing the main updates and the benchmark comparisons. [ 3 ]
- The Nano Banana 2 model ID “gemini-3.1-flash-image” is deprecated and its service is scheduled to end. [ 1 ]