Skip to content
Hacker News front page

C2C lets LLMs talk via KV-cache, 2.5× faster and 3–5% more accurate than text

Cache-to-Cache: Direct Semantic Communication Between Large Language Models

This ICLR'26 paper proposes Cache-to-Cache (C2C), where multiple LLMs communicate by directly exchanging KV-cache instead of generating text. A neural network projects and fuses the source model's KV-cache into the target model, with a learnable gate selecting which layers benefit. C2C beats single models by 6.4–14.2% in average accuracy, outperforms text-based communication by ~3.1–5.4%, and delivers an average 2.5× latency speedup. Code is open at thu-nics/C2C.

Why it matters: Tsinghua and other institutions propose C2C, letting models pass semantics by projecting and merging KV-Cache directly instead of generating text, with 3.1-5.4% higher accuracy and 2.5x lower latency.

Read the original ↗Export Markdown