EdgeRock open-sourced: 72 MiB KV cache at 256K context, 99.1% smaller than Qwen3.5-4B
alifcoder · x · 2026-10-27
rockmatrix-ai released EdgeRock, an open-source multimodal model built for "strong reasoning at constant memory": O(N) generation with O(1) persistent history.
- Instead of growing KV caches, its attention KV cache stays fixed at 72 MiB per sequence even at 256K context — 99.1% smaller than Qwen3.5-4B's 8 GiB
- Fixed-capacity Rock Memory Anchors (RMA) retrieve selected information from earlier in the context even after it leaves the local attention window
- Ships with Apache-2.0 inference/eval code on GitHub (650 stars), Hugging Face weights, an Arxiv paper and whitepaper
Related event: EdgeRock Open-Sources Model with Constant 72 MiB KV Cache at 256K Context(2 posts)→
More from Infra
- Anthropic hits $16.3B H1 revenue as compute efficiency jumps 5.2x in 18 months — daniel_mac8 · 2026-10-27
- Tenstorrent runs Proteina protein model at high throughput, claims no HBM shortage — gajesh · 2026-10-27
- YC-backed Hopper launches voice-agent inference cloud: 80 ms TTFT at $0.30/1M tokens — ycombinator · 2026-10-27
- A100 rental prices still rising in 2026 per a16z data, Jevons paradox hits compute — VraserX · 2026-10-27
- Anthropic S-1 reveals $800B+ in off-books future compute commitments — GaryMarcus · 2026-10-27
- ~42% of Anthropic's ~11GW contracted compute is NVIDIA GPUs — ns123abc · 2026-10-27