Yandex Turns KV Cache into an Agent Runtime for Real-Time LLM Interaction

Yandex Research shows that treating the KV cache as an editable agent runtime—without touching model weights—gives LLMs real-time interactivity, demonstrated by Qwen3.8 playing DOOM interactively.

2026-09-07 ~ 2026-09-07 · 3 related posts

• Yandex researchers propose KV cache as an agent runtime, demo Qwen3.8 playing DOOM interactively — _puhsu · 2026-09-07
• Yandex Makes Pretrained LLMs Interactive by Reorganizing KV Cache, No Retraining Needed — arpit_bhayani · 2026-09-07
• Yandex Publishes 'The KV Cache as an Agent Runtime': Inference-Time Interactivity for LLMs — arpit_bhayani · 2026-09-07