Nano Banana 2.1 Costs About a Third of Pro's 2K Price, but No Clear Winner for Text-Heavy Images

The latency figures come from a script that times the whole round trip through apimodels.app. It posts a generation request, gets a task id back, then sleeps 4 seconds between status checks until the task leaves the pending or processing state ^[15]. The clock starts before the first request ^[15]. Every timing therefore includes the platform's queue. Each one can also overshoot the real finish by up to one poll interval. On the 25-second infographic run, the overshoot can reach 16 percent of the reading ^[20].

The ratios hold up despite that noise. Pro took about 2.3 times as long as 2.1 on the infographic and 1.75 times as long on the edit ^[17]. The post reports one timing per prompt ^{[1] [3]}. For those ratios to carry over to another team's workload, its requests would need similar queue depth at the same 2K, 16:9 settings ^[1].

On quality, the two prompts came out even. Both models spelled every string in the infographic correctly. Pro drew richer icons and 2.1 went flatter ^[5]. The author wrote that for text-heavy graphics at this size, they could not say which one is better without a style preference ^[2]. The edit's reference image was a title card from the author's own site, so the brand name sits on the TV screen ^[6]. As product placement goes, it is at least disclosed.

Price is the figure that transfers, because it comes from Google's pricing page, updated 7 October, at the standard tier ^[10]. Google's docs call 2.1 "the more efficient counterpart to Gemini 3 Pro Image" ^[8]. Pro costs 2.7 times as much at 2K and four times as much at 1K ^[11]. Turned around, 2.1 costs about 37 percent of Pro at 2K and 25 percent at 1K ^[18]. The author's summary range of "half to a third" ^[21] understates the 1K gap. The post does not print the 4K ratio.

The method deserves credit. The author fixed the criteria before looking at any output ^[10] and disclosed the apimodels.app affiliation up front ^[9]. The post also says "Two prompts prove nothing about reliability" ^[4]. That warning applies to 2.1's production record too. All 11 of Pro's failures were content-moderation refusals ^[12]. 2.1's single failure came from upstream, and the author asks readers to treat its percentiles as a first look ^[13].

Google's list price is the one hard number in the post. Text rendering showed no gap on the one test that measured it ^[1]. In that context I think 2.1 is the right default, with Pro as an upgrade chosen scene by scene ^[22]. The four-versus-five limit applies to character references. The post separately says 2.1 accepts up to 10 references on apimodels.app ^[16].