Developer of the Kimi assistant, long known for long-context handling.
Website ↗Kimi resets usage for all subscription members within their validity period, compensating 25% of the monthly total (about 1 week of usage); any shortfall below 25% is reset to 0.Machine-translated
Kimi 就会员使用 K3.1 Pro 出现限流等问题致歉，为所有订阅会员进行重置
Kimi 为所有有效期内的订阅会员重置用量，补偿月度总额的25%（约一周用量），不足25%的重置为0。
The model inference API charges on a pay-as-you-go basis for both Input and Output. For the K3 series models, cache writes are billed separately based on TTL (5min / 1h); input that hits the cache is billed only at the cache-hit price, and the cache write fee is not charged again. If you upload and extract document content and transmit the extracted document content to the model as Input, then the document content will also be billed on a pay-as-you-go basis. File-related APIs (file content extraction / file storage) are free for a limited time, meaning that when you only upload and extract documents, the file APIs themselves incur no fees. For details on how to improve cache hit rates and choose the appropriate TTL tier, see Context Caching Best Practices.Machine-translated
Kimi 文件内容抽取与存储接口限时免费
模型推理接口对 Input 和 Output 均实行按量计费。对于 K3 系列模型，缓存写入按 TTL（5min / 1h）单独计费；缓存命中的输入仅按缓存命中价格计费，不再重复收取缓存写入费用。如果您上传并抽取文档内容，并将抽取的文档内容作为 Input 传输给模型，那么文档内容也将按量计费。文件相关接口（文件内容抽取/文件存储） 限时免费 ，即您只上传并抽取文档时，文件接口本身不产生费用。关于如何提高缓存命中率、选择合适的 TTL 档位，详见 上下文缓存最佳实践 。
Rankings from Artificial Analysis, fetched 2026-11-03 22:43.
