2026-11-03 · America/Los_Angeles · 社区动态 · #5

Prompt caching patterns across major LLM providers look very different

If anyone's confused by the graph. From what I understand the dark green is cached tokens (basically reused input), grey is uncached tokens (new input being processed) and orange is output tokens (what the model generates). Thought I'd give you guys some context. I find it interesting how much higher Anthropic and Cursor are on cached tokens compared to Google and apparently caching can also give some indication of how heavily models are being used for agent workflows since agents tend to reuse a lot of context. Full article: https://ramp.com/data/googles-models-lag-in-agentic-ai-usage

热度 46.0 / 100；排名与评分保留该期记录。