Fetching from the wire…
Skills2026-03-19 · source-backed
Stores KV cache in a radix tree enabling automatic reuse when requests share common prefixes (system prompts, tool definitions). For agentic pipelines with shared context, eliminates redundant computation. February 2026 results show 25x on NVIDIA GB300 NVL72. Drop-in vLLM replacement via OpenAI-compatible server. Markaicode
Each link below shares sources, entities, or timing with this story.
ChatGPT built by OpenAI / Shared entities / What happened next
Linked by a graph relationship (ChatGPT built by OpenAI); both cover February, OpenAI; picks up the February thread on 2026-08-17.
OpenAI partners with Google / Shared entities / What happened next
Linked by a graph relationship (OpenAI partners with Google); both cover February, OpenAI; picks up the February thread on 2026-07-29.
Linked by a graph relationship (OpenAI partners with Google); both cover February, OpenAI; picks up the February thread on 2026-07-25.
Drew Pusateri works at OpenAI / Shared entities / What happened next
Linked by a graph relationship (Drew Pusateri works at OpenAI); both cover February, OpenAI; picks up the February thread on 2026-07-10.
OpenAI released Frontier / Shared entities / What happened next
Linked by a graph relationship (OpenAI released Frontier); both cover February, OpenAI; picks up the February thread on 2026-07-08.
OpenAI released ChatGPT Enterprise / Shared entities / What happened next
Linked by a graph relationship (OpenAI released ChatGPT Enterprise); both cover February, OpenAI; picks up the February thread on 2026-06-23.
OpenAI released Lockdown Mode / Shared entities / What happened next
Linked by a graph relationship (OpenAI released Lockdown Mode); both cover February, OpenAI; picks up the February thread on 2026-06-07.
OpenAI partners with Google / Shared entities / What happened next
Linked by a graph relationship (OpenAI partners with Google); both cover February, OpenAI; picks up the February thread on 2026-05-20.