谷歌Nano Banana 2.1发布：性能涨了，API反而半价 谷歌近日正式发布了Nano Banana 2.1，这一版本在性能上实现了显著提升，但令人意外的是，其API价格却直接砍半。对于开发者和企业用户来说，这无疑是一个“加量还降价”的利好。 性能方面，Nano Banana 2.1在推理速度、准确率和多模态处理能力上均有明显进步。官方数据显示，相比上一代，新版本在典型任务中的响应速度提升了约40%，同时保持了更低的资源占用。这意味着在相同硬件条件下，可以支撑更高的并发请求，或者用更少的算力完成同样的工作。 更引人注目的是定价策略。谷歌将Nano Banana 2.1的API调用价格下调了50%，几乎是对半砍。在AI模型服务普遍按token或调用次数计费的当下，这种“性能涨、价格降”的组合并不常见。分析认为，谷歌此举意在加速抢占开发者生态，尤其是在中小团队和独立开发者市场，低价策略能有效降低试错成本，吸引更多人接入。 对于已经使用上一代API的用户，迁移到2.1版本预计无需大幅改动代码，兼容性较好。谷歌也提供了详细的迁移指南和价格对比工具，方便用户评估成本变化。 总体来

Beating AI Brief: Google has officially released Nano Banana 2.1, an upgraded version of Nano Banana 2 that continues to position itself as a highly efficient image generation and editing model. The main improvements are focused on image quality, instruction following, text generation, and multi-turn character consistency. Gemini, AI Mode, AI Studio, Flow, Stitch, Google Ads, and other products have already begun rolling it out, with the API model name gemini-nano-banana-2.1.

Image editing is the focus this time. Version 2.1 supports 1K, 2K, and 4K, can use up to 14 reference images simultaneously, and maintains consistency for up to 4 characters and 10 objects. Google has also strengthened mask editing, meaning only specified areas are changed, and improved infographic layout, text in images, and ultra-wide aspect ratio generation. The model can now also choose among three thinking levels: minimal, medium, and high.

In Google's own model card, 2.1 with thinking enabled has higher point estimates than Nano Banana 2 and Nano Banana Pro across all 10 evaluations, including text-to-image, general editing, multi-person consistency, and mask editing. The independent Arena leaderboard also confirms a clear improvement: 2.1 currently ranks 5th in text-to-image and 6th in single-image editing, the highest among Google models, but it still trails GPT Image 2.5 and GPT Image 2.

The price has actually dropped. Standard API image output for 1K, 2K, and 4K is $0.0336, $0.0504, and $0.0756 respectively, roughly only half that of Nano Banana 2. However, the input price has risen from $0.50 per million tokens to $1.50, and text and thinking output has also increased from $3 to $7.50. The more reference images and the heavier the thinking, the smaller the cost advantage brought by the lower image prices.