#Telegram

1 messages · Page 1 of 1 (latest)

ebon temple
#

If the LLM rejected in telegram, can change the llm from anthropoc to openai or something else from telegram

serene rampartBOT
lofty laurelBOT
#

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

serene rampartBOT
tropic fulcrumBOT
#

Yes, if that model/provider is already configured on the gateway.

From Telegram, send a standalone command like:

/model
/model list
/model openai/gpt-5.4
/model status

So if Anthropic is rejecting, you can switch the current session to OpenAI from Telegram itself.

Two gotchas:

  • the target model must be available/authenticated on that OpenClaw instance
  • if agents.defaults.models is set, you can only pick from that allowlist
ebon temple
#

Still getting LLM request rejected

tropic fulcrumBOT
#

“LLM request rejected” is a bit too generic, so first please paste:

openclaw --version

and the full rejection text from Telegram.

One important gotcha though: if you switched models inside the same broken chat/session, that may not help. If the rejection is:

LLM request rejected: messages.content.tool_use.input field required

that usually means the session history is stale/corrupted, and the fix is to start fresh:

# 
/new
/model openai/gpt-5.4

So the order is:

  1. /new
  2. switch model if you want
  3. retry

If it’s instead an Anthropic rate limit/billing rejection, then /model openai/gpt-5.4 is the right move, but OpenAI must already be configured on that gateway.

Relevant docs:

#

Paste the exact error and version, and I’ll tell you which branch you’re hitting.

ebon temple
#

🦞 OpenClaw 2026.3.23-2 (7ffe7e4)
🧠 Model: anthropic/claude-opus-4-5 · 🔑 token (anthropic:claude)
🗄️ Cache: 96% hit · 54k cached, 2.0k new
📚 Context: 0/200k (0%) · 🧹 Compactions: 0
🧵 Session: agent:main:main • updated just now
⚙️ Runtime: direct · Think: low
🪢 Queue: collect (depth 0)

tropic fulcrumBOT
#

Nice, that helps.

What this tells me:

  • your version is 2026.3.23-2
  • the current session is still using Anthropic
  • so the switch to OpenAI has not actually taken effect yet

Try this in the Telegram chat itself:

/new
/model openai/gpt-5.4
/model status

What you want to see after that is /model status showing an OpenAI model, not anthropic/claude-opus-4-5.

#

If you paste openclaw models status, I can usually tell pretty quickly which of the three it is.

#

Yep, that error is Anthropic billing/usage, not a Telegram issue.

You're out of extra usage means the current Claude-auth credential has no extra usage left.

You have 4 practical options:

  1. Add Anthropic extra usage

  2. Switch this Telegram session to another model

    /model openai/gpt-5.4
/model status

    or just:

#

If you want, I can help you set up a clean fallback chain next so this auto-recovers instead of dying on Claude usage limits.

ebon temple
tropic fulcrumBOT
#

If you’re still getting the exact Anthropic extra-usage error, then OpenClaw is still on Anthropic. The model switch did not happen.

Please try this exact flow in Telegram:

/new
/model openai/gpt-5.4
/model status

If /model status still shows anthropic/..., then the problem is almost certainly one of these:

  1. OpenAI is not configured/authenticated
  2. OpenAI is blocked by the model allowlist
  3. the /model command is not being applied in that Telegram session

Please paste these next:

# 
openclaw models status
openclaw config get agents.defaults.models

That will tell us immediately whether OpenAI is available to switch to.

If you want, also paste the output of /model status from Telegram after trying the commands above.

ebon temple
tropic fulcrumBOT
#

Ah, I think I see it now.

If you keep sending /model ... in Telegram and still get the same Anthropic extra-usage error, that usually means /model is not being handled as a command at all. It is likely being forwarded to the current model as plain text, and Anthropic rejects it before any switch happens.

In OpenClaw, that happens when either:

  • text commands are disabled (commands.text: true is required), or
  • you are not authorized to use directives like /model, or
  • Telegram native commands are disabled/not registered

Please check on the host:

openclaw config get commands.text
openclaw config get channels.telegram.commands.native
openclaw config get commands.allowFrom
#

Also, in Telegram, try using the native slash-command picker/menu instead of typing it manually. If /model is missing there, that’s another hint native commands are off.

If you just want an immediate workaround, switch the default model from the host:

openclaw models set openai/gpt-5.4

Then retry Telegram.

Docs:

Paste those 3 config values if you want, and I’ll tell you the exact fix.

ebon temple
#

No working, is there a way around it

tropic fulcrumBOT
#

Yes, but the workaround is not inside that broken Telegram session.

Right now the cleanest workaround is to switch the default model from the host:

openclaw models set openai/gpt-5.4
openclaw models status

Then try Telegram again.

If that works, I’d also add a fallback so Anthropic outages/usage limits stop breaking chat:

openclaw models fallbacks add openai/gpt-5.4