Error 400 | Friends of the Crustacean 🦞🤝 | Page 1

strange grotto Apr 14, 2026, 4:57 PM

#

LLM error: {
"error": {
"code": 400,
"message": "Budget 0 is invalid. This model only works in thinking mode.",
"status": "INVALID_ARGUMENT"
}
}

tribal geyserBOT Apr 14, 2026, 4:57 PM

#

modest ingotBOT Apr 14, 2026, 4:57 PM

#

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

tribal geyserBOT Apr 14, 2026, 4:57 PM

#

subtle moonBOT Apr 14, 2026, 4:59 PM

#

That usually means the session is sending thinking off to a model that requires thinking on.

Most likely causes:

you previously set /think off in this thread, or
the agent/config has thinkingDefault: "off" / a model param disabling thinking.

Quick fix:

In this thread, send: /think low
Retry the prompt

If it keeps happening, check config for a disabled thinking default and change it to low (or remove the override so the model default applies).

Also, please paste:

openclaw --version
the exact model id you’re using

Docs:

#

If you want, I can help you find the exact config key next.

strange grotto Apr 14, 2026, 5:36 PM

#

So what do I do? The system is unresponsive other than providing that error

subtle moonBOT Apr 14, 2026, 5:37 PM

#

Do this first:

Send exactly this in the thread:
/think low
Then send your prompt again.

That error means OpenClaw is currently telling the model to use thinking off (budget 0), but that model refuses to run without thinking.

If /think low doesn’t fix it, try this next:

start a new thread/session
send /think low as the first message
then retry

If you want the permanent fix, set a non-off default in config and restart:

#

{
  agents: {
    defaults: {
      thinkingDefault: "low",
    },
  },
}

Also check for any model params that explicitly disable thinking and remove those.

If it’s still stuck after that, paste:

openclaw --version
your model id
whether you’re using chat, Control UI, or CLI

#

Docs:

strange grotto Apr 14, 2026, 5:40 PM

#

What does thinking low change? I don’t want the system to operate poorly I’m wondering if this is an error that hit when I changed LLM’s from Claude to Gemini

subtle moonBOT Apr 14, 2026, 5:41 PM

#

Yep, that very likely started when you switched from Claude to Gemini.

thinking low does not mean “bad quality mode”.
It means:

off = no reasoning budget
low = small reasoning budget
medium/high = more reasoning, usually slower + more expensive

So the jump is really from broken to working, not from “good” to “bad”.

What probably happened:

your session/config was effectively on thinking off
Gemini got budget 0
that Gemini model only accepts thinking on
so it hard-failed with that error

#

My recommendation:

use /think low to get it unstuck
if you’re doing heavier tasks later, bump to /think medium or /think high

So yes, this smells like a Claude → Gemini behavior mismatch, not your whole system degrading.

If you want, I can help you pick a sane default:

low for general use
medium for more agentic work
high only when you really want deeper reasoning

strange grotto Apr 14, 2026, 5:48 PM

#

Now says: ⚠️ The AI service is temporarily overloaded. Please try again in a moment.

subtle moonBOT Apr 14, 2026, 5:49 PM

#

That’s actually a different, better error.

It means the thinking-budget problem is gone, and now you’re just hitting a temporary provider overload / rate-limit.

What to do:

wait 30 to 60 seconds
retry
if it keeps happening, switch back to the Claude setup you had before, or try a less-busy Gemini option

So at this point:

Budget 0 invalid = config/session mismatch
AI service temporarily overloaded = provider-side traffic problem

If this keeps repeating, paste:

openclaw --version
the exact Gemini model you switched to

#

Then I can help narrow down whether it’s just provider load or a bad model/config combo.

strange grotto Apr 14, 2026, 6:09 PM

#

OpenClaw 2026.4.12 (1c0672b)

subtle moonBOT Apr 14, 2026, 6:10 PM

#

Perfect, thanks, that version is recent enough that I wouldn’t treat this as “you’re on an ancient broken build.”

So my read now is:

Version is probably fine
this is more likely a Gemini + thinking-setting mismatch
and the later “overloaded” message was just a provider-side transient

What I’d do next:

check the current session setting with:
/think
if it says off, set:
/think low
if you want this to stop happening after model switches, set a default like:

#

{
  agents: {
    defaults: {
      thinkingDefault: "low",
    },
  },
}

then restart the gateway.

Also, if you can, send the exact Gemini model id you switched to, because some Gemini models are much pickier about thinking being enabled.

So, short version:

No, 2026.4.12 doesn’t look like the main problem
Yes, switching from Claude to Gemini is still the most likely trigger

strange grotto Apr 14, 2026, 6:12 PM

#

Right now, I have it set to think high

subtle moonBOT Apr 14, 2026, 6:14 PM

#

Ah, that’s useful.

If the session is really on think high, then this should not be happening. So I don’t think the chat-level thinking setting is the culprit anymore.

At that point I’d suspect one of these:

stale session state from before the Claude → Gemini switch
a per-model config override for the Gemini model, like:
```
params: { thinking: "off" }
```
a Gemini-side/OpenClaw bug on that model path

Best next test:

start a fresh thread/session
switch to the Gemini model again

#

send /think high
then try a simple prompt

If the fresh session still throws Budget 0 is invalid, then I’d want:

the exact Gemini model id
the relevant agents.defaults.models[...] config block for that model (redact secrets)

Because if /think is already showing high, then something lower-level is still sending budget 0, and that starts looking bug-ish rather than user error.

strange grotto Apr 14, 2026, 6:32 PM

#

Aistudio shows no rate limit issues
Plenty of money in the account
Maybe I need help switching to Claude so I can work again
Then try to switch to Gemini 3.1 pro

subtle moonBOT Apr 14, 2026, 6:33 PM

#

Yeah, I’d unblock yourself first.

Get back onto Claude now

In this thread, do:

/model list
pick the Claude entry that was working before, or if you have the alias:
/model opus
or
/model sonnet
keep thinking enabled, for example:
/think high

If the alias doesn’t exist in your config, /model list is the safest route.

Then try Gemini again, cleanly

#

I’d do that in a fresh thread/session, not this one.

switch to Gemini 3.1 Pro
- likely /model gemini if that alias exists
- otherwise use /model list
immediately set:
/think high
then try a simple prompt

If fresh thread + Gemini + think high still throws Budget 0 is invalid, then I’d treat that as a likely bug/config issue, not user error.

Docs:

If you want, next paste the output of /model list and I’ll tell you the exact line to pick.

strange grotto Apr 14, 2026, 6:34 PM

#

“Failed to set model”

subtle moonBOT Apr 14, 2026, 6:35 PM

#

Got it, that’s a separate problem from the Gemini thinking error.

Failed to set model usually means one of these:

the alias you used doesn’t exist
that model isn’t in your allowlist/config
auth for that provider/model isn’t available

Try this instead:

/model list
choose the Claude model by number, not alias
- example: /model 2
if that still fails, run:
/model status

If the Claude model you want does not appear in /model list, then it isn’t currently configured/allowed.

#

If it does appear but still won’t switch, paste:

what exact /model ... command you ran
the output of /model status

That should tell us whether this is an alias/config/auth issue.

strange grotto Apr 14, 2026, 6:36 PM

#

it just says failed to set model: Gateway request error: model not allowed: Google/status

subtle moonBOT Apr 14, 2026, 6:37 PM

#

Ah, that helps.

Two things are happening:

Google/status is being parsed like a model name, not the status subcommand.
So use exactly:
/model status
More importantly, /model can only switch to models that are in your allowed/configured model list.

So if Claude is not in /model list, you won’t be able to switch to it from chat until it’s added to config.

A sane config looks like this:

#

  agents: {
    defaults: {
      model: { primary: "anthropic/claude-opus-4-6" },
      models: {
        "anthropic/claude-opus-4-6": { alias: "opus" },
        "google/gemini-3.1-pro-preview": { alias: "gemini" },
      },
    },
  },
}

Then restart the gateway:

#

openclaw gateway restart

After that, in chat:

/model opus
/model gemini

So the immediate takeaway is:

this is now mostly an allowlist/config problem
not just a Gemini outage

If you paste your models block, I can tell you exactly what to add.

strange grotto Apr 14, 2026, 6:38 PM

#

I entered model status and model list and I got the same bill for that model error

subtle moonBOT Apr 14, 2026, 6:39 PM

#

Yeah, at that point I’d stop fighting the in-chat /model commands for now.

If /model status and /model list are also tripping the same path, the fastest recovery is to set Claude in config directly and restart.

Use this route:

Open your config file:
~/.openclaw/openclaw.json

If you’re not sure that’s the active file, run:
openclaw health --verbose
(it prints the config path)
Make sure your model block includes both Claude and Gemini, with Claude as primary for now:

#

{
  agents: {
    defaults: {
      model: { primary: "anthropic/claude-opus-4-6" },
      models: {
        "anthropic/claude-opus-4-6": { alias: "opus" },
        "google/gemini-3.1-pro-preview": { alias: "gemini" },
      },
      thinkingDefault: "high",
    },
  },
}

Restart the gateway:

#

openclaw gateway restart

That should get you back onto Claude so you can work again.

Then later, once both models are in the allowlist, switching should be much cleaner.

If you want, paste your current agents.defaults block and I’ll rewrite it exactly.

strange grotto Apr 14, 2026, 6:41 PM

#

I don’t see this: ~/.openclaw/openclaw.json

#Error 400

Get back onto Claude now

Then try Gemini again, cleanly