Gemini model not working | Friends of the Crustacean 🦞🤝 | Page 1

tawdry hazel Apr 17, 2026, 3:32 AM

#

gemini-3-pro-prievew is failing, I use it via OAuth. It has been my most reliable model until recently. Does it not work anymore?

brittle shoreBOT Apr 17, 2026, 3:32 AM

#

north flowerBOT Apr 17, 2026, 3:32 AM

#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

brittle shoreBOT Apr 17, 2026, 3:32 AM

#

sly chasmBOT Apr 17, 2026, 3:35 AM

#

It should still work, so I don’t think Gemini support was removed outright.

The first thing I’d check is which Gemini path you’re actually using, because “OAuth Gemini” can mean different providers in OpenClaw:

google/* = API key path, not OAuth
google-gemini-cli/* = Gemini CLI OAuth
google-antigravity/* = a different OAuth bridge

So if your model ref starts with google/..., that’s already a clue.

Can you paste these?

openclaw --version
openclaw models status
openclaw status --all

#

Also paste the exact failing model ref and the exact error. Preview model IDs are strict, so even a small typo can break it.

One known gotcha for Gemini CLI OAuth: if login succeeded but requests started failing, set one of these on the gateway host and retry:

export GOOGLE_CLOUD_PROJECT=your-project-id
# or
export GOOGLE_CLOUD_PROJECT_ID=your-project-id

Docs:

If you paste the outputs, I can narrow down pretty quickly whether this is:

wrong provider path,
expired/bad OAuth state,

#

a preview model change upstream, or
an OpenClaw version-specific bug.

tawdry hazel Apr 17, 2026, 3:51 AM

#

jon@hp-server:~$ openclaw --version
OpenClaw 2026.4.15 (041266a)

sly chasmBOT Apr 17, 2026, 3:51 AM

#

Thanks, that version is recent enough, so this probably isn’t just “you’re on an ancient build.”

Also, gemini-3-pro-preview is still present in current OpenClaw source for the Gemini CLI OAuth path, so it does not look intentionally removed.

Next thing I need is:

openclaw models status

and ideally the exact runtime error from:

openclaw logs --follow

Two quick sanity checks too:

#

Make sure the model ref is actually spelled preview, not prievew
If that typo made it into config, that alone would break it.
If you’re using Gemini CLI OAuth, the provider should usually be google-gemini-cli/..., not google/...

If requests are failing after login, also try setting one of these on the gateway host:

export GOOGLE_CLOUD_PROJECT=your-project-id
# or
export GOOGLE_CLOUD_PROJECT_ID=your-project-id

Paste openclaw models status and the error text, and I can narrow it down pretty fast.

tawdry hazel Apr 17, 2026, 3:54 AM

#

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (3): anthropic (1), google-gemini-cli (1), openai-codex (1)

anthropic effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=1, api_key=0) | anthropic:default=token:sk-ant-o...izy4LrRG
google-gemini-cli effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=1, token=0, api_key=0) | google-gemini-cli:jon9314@gmail.com=OAuth (jon9314@gmail.com)
ollama effective=env:ol...al | env=ol...al | source=env: OLLAMA_API_KEY
openai-codex effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=1, token=0, api_key=0) | openai-codex:default=OAuth
openrouter effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | openrouter:default=sk-or-v1...6b3c58fa | env=sk-or-v1...6b3c58fa | source=env: OPENROUTER_API_KEY
perplexity effective=env:sk-or-v1...6b3c58fa | env=sk-or-v1...6b3c58fa | source=env: OPENROUTER_API_KEY

OAuth/token status
[openai-codex] Token refresh failed: 401 {
"error": {
"message": "Your refresh token has already been used to generate a new access token. Please try signing in again.",
"type": "invalid_request_error",
"param": null,
"code": "refresh_token_reused"
}
}
[openai-codex] Token refresh failed: 401 {
"error": {
"message": "Your refresh token has already been used to generate a new access token. Please try signing in again.",
"type": "invalid_request_error",
"param": null,
"code": "refresh_token_reused"
}
}

anthropic
- anthropic:default static
google-gemini-cli usage: Pro 100% left · Flash 100% left
- google-gemini-cli:jon9314@gmail.com (jon9314@gmail.com) ok expires in 0m
openai-codex
- openai-codex:default ok expires in 0m

sly chasmBOT Apr 17, 2026, 3:55 AM

#

This is helpful. Your Gemini OAuth itself looks fine.

What I see:

google-gemini-cli auth profile is present and healthy
usage shows Pro 100% left / Flash 100% left
so this does not look like “Gemini OAuth stopped working”
the openai-codex refresh_token_reused errors are unrelated noise here

So the likely problem is one of these:

the exact model ref
a Gemini CLI request/runtime error
missing GOOGLE_CLOUD_PROJECT / GOOGLE_CLOUD_PROJECT_ID
local gemini CLI version skew

Can you paste these next?

#

openclaw models list --provider google-gemini-cli
gemini --version
openclaw logs --follow

Then send one prompt with the failing model and grab the exact Gemini error from the log.

Also, small but real thing: if the actual configured ref is spelled prievew instead of preview, that would absolutely fail.

My read so far: auth is okay, the failing part is probably the specific gemini-3-pro-preview request path, not Gemini OAuth as a whole.

tawdry hazel Apr 17, 2026, 3:55 AM

#

jon@hp-server:~$ openclaw models status
Config warnings:\n- plugins.allow: plugin not found: exec (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.15 (041266a) — Give me a workspace and I'll give you fewer tabs, fewer toggles, and more oxygen.

Config warnings:\n- plugins.allow: plugin not found: exec (stale config entry ignored; remove it from plugins config)
Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : google-gemini-cli/gemini-3-pro-preview
Fallbacks (9) : google-gemini-cli/gemini-3-flash-preview, ollama/gemma4:e4b, openai-codex/gpt-5.1, openai-codex/gpt-5.3-codex-spark, openai-codex/gpt-5.3-codex, openai-codex/gpt-5.2, openrouter/google/gemma-3-4b-it, openrouter/meta-llama/llama-3.1-8b-instruct:nitro, openrouter/deepseek/deepseek-v3.2
Image model : -
Image fallbacks (0): -
Aliases (0) : -
Configured models (24): openrouter/deepseek/deepseek-v3.2, openai-codex/gpt-5.1, openai-codex/gpt-5.3-codex, google-gemini-cli/gemini-3-pro-preview, google-gemini-cli/gemini-3-flash-preview, openai-codex/gpt-5.3-codex-spark, openai-codex/gpt-5.2, openrouter/hunter-alpha, google-gemini-cli/gemini-1.5-flash, google-gemini-cli/gemini-1.5-pro, google-gemini-cli/gemini-2.0-flash, google-gemini-cli/gemini-2.0-pro-exp, google-gemini-cli/gemini-2.5-flash, google-gemini-cli/gemini-2.5-pro, openai-codex/gpt-4o, openai-codex/gpt-4o-mini, openai-codex/o1, openai-codex/o1-mini, openai-codex/o3-mini, openai-codex/gpt-5.4, anthropic/claude-3-5-sonnet-latest, anthropic/claude-3-opus-latest, anthropic/claude-3-haiku-latest, ollama/kimi-k2.5:cloud

sly chasmBOT Apr 17, 2026, 3:56 AM

#

Nice, this narrows it down a lot.

google-gemini-cli/gemini-3-pro-preview is:

configured
allowed
selected as default
backed by a healthy OAuth profile

So no, this does not look like “OpenClaw no longer supports that model.”

At this point the remaining suspects are mostly on the Gemini CLI/runtime side, not OpenClaw config:

the local gemini CLI changed behavior
that specific model is flaky/upstream-broken right now
missing GOOGLE_CLOUD_PROJECT / GOOGLE_CLOUD_PROJECT_ID
a runtime error only visible in logs

#

Two things I’d do next:

gemini --version
openclaw logs --follow

Then send one message through OpenClaw with the failing model and paste the exact error line.

Also, the warning below is probably unrelated, but worth cleaning up later:

plugins.allow: plugin not found: exec

If you want a fast A/B test, temporarily switch once to:

#

google-gemini-cli/gemini-2.5-pro

If that works while google-gemini-cli/gemini-3-pro-preview fails, I’d strongly suspect an upstream Gemini CLI/model issue rather than broken OAuth.

tawdry hazel Apr 17, 2026, 3:58 AM

#

jon@hp-server:~$ openclaw models list --provider google-gemini-cli
Config warnings:\n- plugins.allow: plugin not found: exec (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.15 (041266a) — Hot reload for config, cold sweat for deploys.

Config warnings:\n- plugins.allow: plugin not found: exec (stale config entry ignored; remove it from plugins config)
Model Input Ctx Local Auth Tags
google-gemini-cli/gemini-3-pro-preview text+image 1024k no yes default,configured
google-gemini-cli/gemini-3-flash-preview text+image 1024k no yes fallback#1,configured
google-gemini-cli/gemini-1.5-flash - - - - configured,missing
google-gemini-cli/gemini-1.5-pro - - - - configured,missing
google-gemini-cli/gemini-2.0-flash text+image 1024k no yes configured
google-gemini-cli/gemini-2.0-pro-exp - - - - configured,missing
google-gemini-cli/gemini-2.5-flash text+image 1024k no yes configured
google-gemini-cli/gemini-2.5-pro text+image 1024k no yes configured
jon@hp-server:~$

sly chasmBOT Apr 17, 2026, 3:58 AM

#

Yep, that answers the “is it gone?” part pretty clearly:

No, it has not been removed from your OpenClaw setup.

Your output shows google-gemini-cli/gemini-3-pro-preview is:

present in the provider catalog
auth-backed
configured
selected as default

So this is not a missing-model or broken-OAuth problem.

At this point I’d treat it as a runtime/model-specific failure until proven otherwise.

My recommendation:

#

Temporary workaround: switch to google-gemini-cli/gemini-2.5-pro for now
Confirmation step: run openclaw logs --follow, send one failing prompt, and paste the Gemini error
Optional but useful: paste gemini --version

The plugins.allow: plugin not found: exec warning looks unrelated.

So, short version: the model still exists, but the request path for that specific Gemini CLI model may be broken/flaky upstream right now.

tawdry hazel Apr 17, 2026, 4:00 AM

#

jon@hp-server:~$ gemini --version
(node:3428998) [DEP0040] DeprecationWarning: The punycode module is deprecated. Please use a userland alternative instead.
(Use node --trace-deprecation ... to show where the warning was created)
0.21.3
jon@hp-server:~$

sly chasmBOT Apr 17, 2026, 4:00 AM

#

Aha, I think we found the likely culprit.

gemini --version = 0.21.3 is a strong clue. I checked, and there’s a Gemini CLI bug report specifically on 0.21.3 where gemini-3-pro-preview access/selection breaks. Current Gemini CLI docs also say to upgrade to latest for Gemini 3 support.

So this now looks much more like a Gemini CLI issue underneath OpenClaw, not an OpenClaw auth/model config issue.

I’d try this:

# npm install
npm install -g @google/gemini-cli@latest

# or Homebrew
brew upgrade gemini-cli

Then:

#

open Gemini CLI directly
/settings → make sure Preview Features = true
restart Gemini CLI completely
restart OpenClaw gateway too

Relevant refs:

Also, there are upstream reports that gemini-3-pro-preview can be overloaded/unreachable at times, so even after upgrading I’d keep google-gemini-cli/gemini-2.5-pro as a fallback.

So my best answer now is: yes, it should still exist, but your local Gemini CLI version is probably the real problem.

If updating fixes it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow can index it.

#Gemini model not working