**Gemini CLI on Windows Terminal — response generation extremely slow, looking for a real fix**
Hi all,
Posting here because I’ve spent quite a few hours troubleshooting this and I’m running out of ideas. Hoping someone in the community (or from the Gemini CLI team) can shed some light.
**Issue summary**
When using Gemini CLI on Windows Terminal, the response generation is consistently slow. The elapsed timer runs for 30 seconds to several minutes per prompt, even for trivial questions. The CLI sometimes shows “Slow response times detected. Automatically switching from gemini-2.5-pro to gemini-2.5-flash for faster responses” or “This is taking a bit longer, we’re still on it…”
This is not a one-off. It happens on:
- Fresh CLI sessions with no context
- Empty project directories
- Simple prompts like “hi” or “what is 2+2”
- Both `gemini-2.5-pro` and `gemini-2.5-flash`
**Steps to reproduce**
1. Install latest Gemini CLI on Windows: `npm install -g @google/gemini-cli@latest`
2. Open Windows Terminal
3. Run `gemini` and authenticate with Google OAuth
4. Type any short prompt and press Enter
5. Watch the elapsed timer climb past 20–30 seconds for what should be an instant reply
**Environment**
- Windows 11 Pro 23H2
- Windows Terminal (latest stable)
- Node.js v20.x LTS
- Gemini CLI: latest
- Auth: Google OAuth (free tier)
- No corporate proxy, no VPN, fiber connection
**Things I’ve tried (none fully resolve it)**
- Reinstalling Node and the CLI from scratch
- `/clear` and `/compress` to wipe context
- Disabling all MCP servers
- Disabling telemetry
- Pinning model to `gemini-2.5-flash` instead of leaving it on auto
- Running with `–debug` (logs show long waits on API responses, not on local tooling)
- `/stats` shows API time is the dominant cost — tool time is minimal
**Partial workaround**
Switching from OAuth to a Gemini API key (from aistudio.google.com) noticeably improves things, suggesting OAuth users are sharing a slower / more congested quota pool. But this shouldn’t be the expected experience for free-tier users.
**What I’d like to know**
1. Is there an official fix in progress, and if so, what’s the rough timeline?
2. Is the slowness primarily Windows-specific, or does it affect macOS/Linux equally?
3. For users sticking with OAuth, are there config tweaks the team recommends?
4. Has anyone tried this in WSL and seen meaningfully better performance?
Would really appreciate hearing from anyone who’s solved this on Windows, or from the team about what’s coming. Thanks!