I get this error, but google ai studio documentation shows 3 pro preview is still active

[dusky oriole]

LLM error: {
"error": {
"code": 404,
"message": "This model models/gemini-3-pro-preview is no longer available. Please update your code to use a newer model for the latest features and improvements.",
"status": "NOT_FOUND"

[blazing folio]

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

[ruby plover]

That 404 is coming from Google’s API, so for the API key/endpoint OpenClaw is using, models/gemini-3-pro-preview is not available even if an AI Studio page still mentions it.

A few likely causes:

• You have an older OpenClaw config/alias pinned to that exact preview model.
• AI Studio UI/docs and the public Gemini API endpoint can lag or differ in model IDs.
• Preview model availability can change by project/region/API surface.

First thing: what does openclaw --version show?

Then check what model OpenClaw is actually resolving with:

openclaw models status
openclaw models list

If you manually configured google/gemini-3-pro-preview, change it to a currently listed Google model from your own openclaw models list output, or remove the pin and let the current OpenClaw alias/config resolve it. The current OpenClaw docs no longer point the gemini shorthand at google/gemini-3-pro-preview.

[ruby plover]

Docs:
https://docs.openclaw.ai/concepts/models
https://docs.openclaw.ai/gateway/configuration

For fresh model picks, #models is the best place since Google preview availability moves fast.

[dusky oriole]

version is OpenClaw 2026.4.12 (1c0672b)

[ruby plover]

Thanks. 2026.4.12 is old enough that I’d treat this as a stale model alias/config issue first.

Current OpenClaw docs point the Gemini shorthand at google/gemini-3.1-pro-preview, not google/gemini-3-pro-preview. So the fastest path is:

openclaw update
openclaw doctor
openclaw gateway restart
openclaw models status
openclaw models list

Then make sure your configured primary model is either the gemini alias after updating, or an exact model ID shown by openclaw models list. If your config explicitly says google/gemini-3-pro-preview, replace/remove that pin.

Docs:
https://docs.openclaw.ai/install/updating
https://docs.openclaw.ai/concepts/models

[ruby plover]

If it still 404s after updating, paste the redacted openclaw models status output and the model-related config snippet.

[dusky oriole]

ok im working on updating it now, but i get this message with a yes or no option, what does this do?

Some skills are allowed for this agent but are not usable in the current runtime environment.

[ruby plover]

That warning is about OpenClaw skills, not the Gemini model error.

Skills are optional capability packs. The message means your config/agent allowlist includes some skills, but this runtime can’t actually use them right now, usually because the OS is wrong, a required binary is missing, or an env/config requirement is missing.

For the yes/no prompt:

• Yes = go through skill setup now. It may offer to install missing dependencies or set env/API-key config for skills.
• No = skip skill setup for now. OpenClaw still updates/runs; those unavailable skills just won’t be injected/usable by the agent.

For your current goal, I’d choose No unless you specifically know you need those skills. Finish the update/model fix first, then you can revisit skills later with:

openclaw doctor
openclaw onboard

Docs: https://docs.openclaw.ai/tools/skills

[dusky oriole]

the yes or no options are for this message "Disable 41 unavailable skills in config?"

[ruby plover]

Ah, that specific prompt is from the skills readiness/fix flow.

Yes writes to your OpenClaw config and sets the unavailable skills to disabled, roughly skills.entries.<skill>.enabled = false. It does not delete the skills or your data, but those skills won’t be offered to agents until you re-enable/configure them.

No leaves your config as-is. You may keep seeing warnings about unavailable skills, but nothing changes.

For this Gemini/model update issue, I’d choose No for now. Disabling 41 skills is a broad config cleanup step, and it’s unrelated to the model 404. Once the model issue is fixed, you can come back and intentionally clean up skills with:

openclaw doctor
openclaw doctor --fix

Docs mention this behavior here: https://docs.openclaw.ai/cli/doctor

[dusky oriole]

ok continuing with the update, can i paste a screenshot of the powershell readout instead of pasting everything? are you able to read images?

[ruby plover]

Yes, I can read screenshots here. A screenshot is fine.

Best version: paste the text output if it’s easy, because commands/errors are easier to search and quote. But if PowerShell output is long or awkward, upload a screenshot and I’ll read it.

Just avoid including secrets: API keys, tokens, full auth headers, private webhook URLs, etc. If they’re visible, crop/blur them first.

[dusky oriole]

alright this is where im at

[ruby plover]

Yep, I can read it. The important part is:

agents.defaults.models.google/gemini-1.5-flash: Unknown model ... Use google/gemini-2.5-flash.

So the update worked, but your config still contains an old Gemini 1.5 Flash model entry. The Discord/voice/brave plugin warnings are separate and not the startup blocker.

Run exactly this first:

openclaw doctor --fix

Then retry:

openclaw gateway restart

[ruby plover]

If doctor --fix does not repair it, open this file:

notepad $env:USERPROFILE\.openclaw\openclaw.json

Find any google/gemini-1.5-flash entry under agents.defaults.models and either remove that old model entry or change it to google/gemini-2.5-flash. Then save and restart the gateway.

This is a config modernization issue now, not the original Gemini 3 Pro preview issue.

[dusky oriole]

why would i change it to 2.5 flash? i want to keep 3 pro preview like i had or whatever the equivalent is

[ruby plover]

Fair question. The 2.5-flash suggestion is only for the old Flash fallback/catalog entry shown in your screenshot:

agents.defaults.models.google/gemini-1.5-flash

That is not the Pro model you’re trying to use. It’s just an obsolete model entry that is making the whole config invalid, so OpenClaw refuses to start.

If you want the current Pro-preview equivalent, don’t change your primary to Flash. Use the current Gemini alias after updating, or explicitly set the Pro-preview model OpenClaw now documents:

agents: {
  defaults: {
    model: "gemini"
  }
}

or, if you prefer the explicit ID:

[ruby plover]

agents: {
  defaults: {
    model: "google/gemini-3.1-pro-preview"
  }
}

The repair you need is: remove/modernize the stale google/gemini-1.5-flash entry under agents.defaults.models. That old Flash entry can exist separately from your primary model, and it’s what’s blocking startup right now.

Docs: https://docs.openclaw.ai/concepts/models

[dusky oriole]

ok back to where im at currently, now im getting this its telling me to use 3.1 pro

Error: Config validation failed: agents.defaults.model.primary: Unknown model: google/gemini-3-pro-preview. Google shut down Gemini 3 Pro Preview on 2026-03-09. Use google/gemini-3.1-pro-preview

[ruby plover]

Yep, that’s the clean answer now: your primary model is still pinned to the old retired ID.

Change:

primary: "google/gemini-3-pro-preview"

to:

primary: "google/gemini-3.1-pro-preview"

or use the shorthand:

[ruby plover]

model: "gemini"

Then restart:

openclaw gateway restart

On Windows, quickest edit path:

notepad $env:USERPROFILE\.openclaw\openclaw.json

So there are two stale entries you’ve hit:

[ruby plover]

• google/gemini-1.5-flash under agents.defaults.models → old catalog/fallback entry, use/remove/fix that separately.
• agents.defaults.model.primary = google/gemini-3-pro-preview → your actual primary, replace with google/gemini-3.1-pro-preview.

That keeps you on the current Pro-preview equivalent, not Flash.

[dusky oriole]

alright im fallig behind a bit here, walk me through the entire process of what i need to do

[ruby plover]

Yep. Do this in order, and don’t worry about the older detours.

Goal: update OpenClaw, remove stale retired Gemini IDs, and set your primary model to the current Pro-preview ID.

1. Open PowerShell.

2. Edit your OpenClaw config:

notepad $env:USERPROFILE\.openclaw\openclaw.json

3. In Notepad, search for:

google/gemini-3-pro-preview

[ruby plover]

Replace it with:

google/gemini-3.1-pro-preview

4. Also search for:

google/gemini-1.5-flash

If you find it under agents.defaults.models, either delete that old model entry or replace it with:

google/gemini-2.5-flash

[ruby plover]

That Flash bit is only cleaning up an old fallback/catalog entry. Your primary Pro model should still be google/gemini-3.1-pro-preview.

5. Save the file and close Notepad.

6. Run config repair/check:

openclaw doctor --fix

If it asks about disabling unavailable skills, you can choose No for now. That’s unrelated to this Gemini fix.

7. Restart the gateway:

openclaw gateway restart

[ruby plover]

8. Verify:

openclaw models status
openclaw --version

If it still errors, send the new screenshot or paste the exact error. The one thing we’re looking for is any remaining retired model ID in ~\.openclaw\openclaw.json.

Docs:
https://docs.openclaw.ai/install/updating
https://docs.openclaw.ai/concepts/models