I got unbanned from gemini-cli on May 8, 2026. I’d not been following the updates to gemini-cli since the ban - a total of almost ~4 weeks. During that time, it looks like gemini-cli has gotten some nice updates to itself. Here are a few that are noteworthy:

  1. Plan Model Routing: when enabled allows the harness to automatically choose a better model
  2. Model seems to default to 3.1-flash-preview to be the primary agent model
  3. Auto theme switching to try and adjust to the terminal background color. It’s not perfect but it’s at least better than having to manually toggle it every time. You can decrease terminal backgroud polling interval but it seems to adjust within a min by default
  4. Use Alternate Screen Buffer is much better to preserve the terminal state before running gemini
  5. New Ink render process for the UI
  6. Incremental Rendering can now be toggled - MAJOR improvements to flickering in the TUI. Kudos gemini-cli team - this was a MUCH needed improvement

Overall, I want to say that I still prefer the gemini models because they jive with how I think. And now that gemini-cli is now accessible again, the total number of frontier cloud tokens have also gone up.

My deep dive into local models still prove out value. I’ve now also figured out how to hot-swap between gemma4 and qwen3.6 as needed for pi.

Solid set of improvements continue further tinkering.