AI News · The field notes
The Daily AI Drop - Morning Edition, November 2, 2026
Why your AI agent costs more after midnight, and how to run big models like DeepSeek and Qwen on a single GPU locally.
Why your AI agent costs more after midnight, and how to run big models like DeepSeek and Qwen on a single GPU locally.
Listen: Listen to this episode
Why your AI agent costs more after midnight — aish_reganti
Prompt caching lets providers reuse work on the unchanged start of your prompt, but only if that prefix stays identical. Coding agents like OpenCode and Kilo Code put today's date in the system prompt, so crossing midnight breaks the cache and forces the model to re-read almost the entire conversation. Other triggers — switching reasoning effort or connecting an MCP server — can do the same, quietly inflating the bill for long agentic sessions.
Colibri Tutorial: Run Big Models Like DeepSeek, Qwen on One GPU Locally — fahdmirza
A hands-on tutorial for Colibri v2.1.0, a tool for running large open models on a single GPU at home. Running models locally keeps your prompts on your machine, avoids per-token charges, and lets you experiment freely. As local tooling improves, the gap between what you can run at home and what you rent in the cloud keeps shrinking, making a hybrid cloud-plus-local setup the normal way serious users work.