Tsinghua Open-Sources C2C: LLMs Communicate via KV-Cache, 2.5x Faster Inference

Tsinghua and Infinigence researchers open-sourced Cache-to-Cache (C2C), accepted by ICLR 2026, which lets LLM agents communicate directly via KV-Cache instead of text tokens, boosting inference speed by 2.5x.

2026-09-18 ~ 2026-09-19 · 2 related posts

• Chinese Researchers Open-Source Cache-to-Cache: LLMs Talk Without Tokens, +14.2% Accuracy — Scobleizer · 2026-09-18
• Tsinghua's C2C Lets LLMs Skip Text and Merge KV-Caches Directly, 2.5x Faster with +14.2% Accuracy — anselm · 2026-09-19