[BUG] Gemini CLI not respecting the set model

[DenisSilent]

What happened?

[ACTION REQUIRED] 📎 PLEASE ATTACH THE EXPORTED CHAT HISTORY JSON FILE TO THIS ISSUE IF YOU FEEL COMFORTABLE SHARING IT.

Not sharing the whole part as I'm not comfortable with it; however this should be enough:

Responding with gemini-3.1-flash-lite-preview

ℹ Switched to fallback model gemini-3.1-flash-lite-preview

ℹ Switched to fallback model gemini-3.1-flash-lite-preview
╭───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ ⊷ Edit soc-working_FIXED copy.ipynb.json: class ResourceManager:... => class ResourceManager:... │
│ │
╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
In Copy Mode. Use Page Up/Down to scroll. Press Ctrl+S or any other key to exit.
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ │
│ Usage limit reached for gemini-3-flash-preview. │
│ Access resets at 10:55 AM GMT+1. │
│ /stats model for usage details │
│ /model to switch models. │
│ /auth to switch to API key. │
│ │
│ │
│ ● 1. Switch to gemini-3.1-flash-lite-preview │
│ 2. Upgrade for higher limits │
│ 3. Stop

---

AKA - on the location, I used this command in Powershell: 'gemini -m 'gemini-3.1-flash-lite-preview' -y --include-directories *' Therefore, explicitly setting Gemini 3.1 Flash Lite as the model I want to use as I already used Gemini 3 Flash limits. However, I keep getting the message above that I should switch to 3.1 Flash Lite after setting it; and even choosing the switch will print that it was switched, but it asks again. Prompting it again, same results. All of that while this is what '/stats' gives me:

╭───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ │
│ Session Stats │
│ │
│ Interaction Summary │
│ Session ID: 436ac121-23e3-4393-ba80-cf93c4a30f2a │
│ Auth Method: Signed in with Google (den.ti@seznam.cz) │
│ Tier: Gemini Code Assist for individuals │
│ Tool Calls: 7 ( ✓ 4 x 3 ) │
│ Success Rate: 57.1% │
│ User Agreement: 100.0% (6 reviewed) │
│ │
│ Performance │
│ Wall Time: 9m 9s │
│ Agent Active: 1m 15s │
│ » API Time: 27.4s (36.4%) │
│ » Tool Time: 47.7s (63.6%) │
│ │
│ Model Reqs Model usage Usage resets │
│ ──────────────────────────────────────────────────────────────────────────────── │
│ gemini-3.1-flash-lite-… 14 ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 26% 10:55 AM (15h 3m) │
│ gemini-3-flash-preview 6 ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ Limit 10:55 AM (15h 3m) │
│ gemini-2.5-flash - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ Limit 10:55 AM (15h 3m) │
│ gemini-2.5-flash-lite - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 26% 10:55 AM (15h 3m) │
│ gemini-2.5-pro - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 36% 1:13 PM (17h 21m) │
│ gemini-3-pro-preview - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 36% 1:13 PM (17h 21m) │
│ │
╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

Therefore, I have a lot of messages for that model left. But it doesn't want to use it. Plus it is probably location-limited, as another session in the same location puts me in this loop again; yet in new location it works.
Therefore, it makes it unusable, and therefore not letting me to use my 1000 messages per day which is propagated on this repo. OAuth login.

What did you expect to happen?

I expected the Gemini CLI to allow me to use it fully and not bug me like this; therefore that it would respect the model choice.

Client information

• CLI Version: 0.36.0-nightly.20260321.fc03891a1
• Git Commit: d99aa4e
• Session ID: 436ac121-23e3-4393-ba80-cf93c4a30f2a
• Operating System: win32 v22.17.0
• Sandbox Environment: no sandbox
• Model Version: gemini-3.1-flash-lite-preview
• Auth Type: oauth-personal
• Memory Usage: 758.5 MB
• Terminal Name: Unknown
• Terminal Background: Unknown
• Kitty Keyboard Protocol: Unsupported

Login information

OAuth login thru personal account; free tier.

Anything else we need to know?

No response

[gemini-cli]

Found possible duplicate issues:

• gemini-3-flash-preview quota prevents me using any other models #23264
• Cannot switch to gemini-2.5-flash after Pro quota limit is reached #11921

If you believe this is not a duplicate, please remove the status/possible-duplicate label.

[kazukinakai]

This is caused by internal utility model configs (loop-detection, llm-edit-fixer, next-speaker-checker, web-search, web-fetch, web-fetch-fallback) being hardcoded to gemini-3-flash-base → gemini-3-flash-preview. Even when you set the main model to Flash Lite, these background calls continue to consume Flash quota and trigger the error.

PR #25684 fixes this by adding gemini-3-flash-lite-base and switching all six configs to use gemini-3.1-flash-lite-preview instead. Flash Lite officially supports all the tools these configs use (including googleSearch and urlContext).