Can same-family models share warm cache? Reddit debates mid-conversation model switching costs
rapPayne · reddit · 2026-11-20
On Reddit, rapPayne cites an edgee.ai article framing mid-conversation LLM switching as a ski-rental problem: every prompt resends full context, and switching models invalidates the warm cache, making 100% of context new and expensive.
His open question: can models within the same vendor family share cache? Since each model has its own tokenizer, could vendors force a shared tokenizer during training or build a common mid-level tokenizer so warm cache survives model switches? No answer yet — an open discussion.
More from coding & agent
- Dev argues agents need heavy investment in skills: a blank session is like an intern with no context — jmschreiber91 · 2026-11-20
- Perplexity adds Decider to Context API to score pages and catch prompt injection — denisyarats · 2026-11-20
- Only 11% of companies have scaled AI agents; the rest pay for AI and still do the work — pritisinghhhh · 2026-11-20
- Antigravity is just a fork of Codex, claims dev citing lag and missing features — jasondeanlee · 2026-11-20
- Slide Maps OpenAI Agents, LangGraph, Mastra and More onto Durable Computation Runtimes — francesc · 2026-11-20
- Designers Spent Years Learning Figma; Now They Build Their Own Tools in Minutes — LukeW · 2026-11-20