EdgeRock Open-Sources Model with Constant 72 MiB KV Cache at 256K Context

rockmatrix-ai has open-sourced EdgeRock, a multimodal model featuring O(1) memory history and O(N) generation complexity. Its attention KV cache stays constant at 72 MiB per sequence at 256K context, 99.1% smaller than Qwen3.5-4B.

2026-10-27 ~ 2026-10-27 · 2 related posts

• EdgeRock-4B holds KV cache at 72 MiB across 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27
• EdgeRock open-sourced: 72 MiB KV cache at 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27