Building the foundation for running extra-large language modelsCloudflare / Apr 16, 2026PD分離で3x高速化キャッシュヒット60%→80%Infireで起動20秒未満prefill-decodekv-cachespeculative-decodinginfiremulti-gpusession-affinity
Workers AI - Moonshot AI Kimi K2.5 now available on Workers AICloudflare Developer Platform / Mar 19, 2026256k コンテキスト対応マルチターンツール呼び出し非同期バッチAPI(プル)workers-aikimi-k2.5prefix-cachingasync-apivision-inputsjson-schemasession-affinity