Today for AI
HOT RADAR
llmHEAT 10.8°

AI CURATED HOT STORY · 10/26/2026

EdgeRock-4B Open-Sourced: Cuts KV Cache by 99%, Boosts Long Inference Throughput 6x

1 reports archived1 independent sourcesupdated 10/26/2026, 14:45:25
Synthesis & Latest Updates
1 Sources Cross-Validated

Yanshan Tech, HKUST, and CMU have open-sourced EdgeRock-4B, a model that reduces KV Cache usage by 99.1% via fixed memory allocation. It maintains constant VRAM at 256K context length, achieving 6.42x throughput on a single GPU while preserving high reasoning accuracy, significantly lowering deployment costs for long-context and agent scenarios.

LATEST/Yanshan Tech, HKUST, and CMU have open-sourced EdgeRock-4B, a model that reduces KV Cache usage by 99.1% via fixed memory allocation. It maintains constant VRAM at 256K context length, achieving 6.42x throughput on a single GPU while preserving high reasoning accuracy, significantly lowering deployment costs for long-context and agent scenarios.

Hourly heat curve

3 fully observed hours
Now
10.8
Peak
11.417:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first
  1. 量子位官网T2·88 pts

    EdgeRock-4B Open-Sourced: Cuts KV Cache by 99%, Boosts Long Inference Throughput 6x

    • KV Cache usage remains constant (72MiB) from 4K to 256K context, reducing memory footprint by 99.1% compared to traditional models
    • Single 80GB GPU supports 512 concurrent long-inference requests with 12,124 tokens/s output throughput, 6.42x higher than the source model
    • Achieves 82.9% average pass@1 on AIME 2024-2026 benchmarks, significantly outperforming other bounded-state models
EdgeRock-4B Open-Sourced: Cuts KV Cache by 99%, Boosts Long Inference Throughput 6x | Today for AI