Two things changed for managed-mode OpenClaw bots on ClawCloud: the model catalog now includes 6 free models, and you can switch to one of them at any point to stop credit usage — including when you're running low.
The problem
Every managed-mode ClawCloud instance gets a monthly credit pool based on the selected credits addon tier — Small ($8), Medium ($25), Large ($60), or XLarge ($100). Credits are the dollar amount of AI API usage your addon covers. When that pool empties, your bot stops responding. The OpenRouter sub-key provisioned for your instance has a hard spending limit, and once it hits that limit, no further API calls go through.
Some users were hitting the limit mid-month, particularly on Lite plans with heavier usage patterns. When they did, the bot went silent until the next billing cycle.
The fix required having something to switch to: free models that work without any API cost.
6 free models in the catalog
OpenRouter's free tier includes open-weight models you can use without any API cost. Those models are now part of the OpenClaw catalog on every managed-mode instance.
Available free models include:
| Model | Provider |
|---|---|
| GPT-OSS 120B | OpenAI |
| Llama 3.3 70B | Meta |
| Mistral Small 3.1 24B | Mistral |
| Qwen3 Coder | Qwen |
| Gemma 3 27B | |
| DeepSeek R1 0528 | DeepSeek |
Send /models to your bot to see the full list.
Free models don't count against your credits at any point in the billing cycle. You can use them by choice now, not just when credits run out.
Switch to a free model before credits run out
The dashboard credit bar turns orange when your usage passes 75%. That's the signal to act.
If you're watching the dashboard and the credit bar has turned orange, and Backup Models is enabled, switch to a supported free model to stop credit accumulation:
/model openai/gpt-oss-120b:free
/model meta-llama/llama-3.3-70b-instruct:free
From that point, no more credits deduct for those conversations. When credits reset at the billing boundary, your bot returns to the paid primary model automatically. No action needed.
The model switching guide covers all ways to change models, including short aliases for paid models and full model IDs for the free tier.
What the credit bar shows
The dashboard credit bar gives you a clear signal for when to act:
- Blue — below 75%, plenty of headroom remaining
- Orange — above 75%, time to consider switching to a free model
- Red — credits exhausted, bot has stopped responding
If the bar turns red, switching to a free model is the fastest way to restore service without waiting for the next billing cycle. See how to keep your bot running when credits run low for step-by-step instructions.
What free models are good for
Free models handle everyday tasks well — answering questions, summarizing content, drafting text, and basic reasoning. They won't match Claude Sonnet or GPT-4.1 on tasks that need strong multi-step reasoning or long-context understanding.
For the kind of usage that typically runs an OpenClaw bot — messages in a Telegram group or Discord server — GPT-OSS 120B handles the routine load without much noticeable difference from paid tiers.
What to do if you consistently hit the limit
Switching to a free model keeps your bot online, but if you're regularly hitting the limit before mid-month, that's worth addressing directly. A few options:
- Upgrade your plan — Pro and Max have larger credit pools.
- Switch to a cheaper primary model — GPT-4.1 Mini and Gemini 2.5 Flash cost far less per message than Sonnet or GPT-4.1 while handling most tasks well.
- Trim your system prompt — the system prompt is sent with every message. A 2,000-token prompt costs tokens on every exchange.
For a breakdown of per-model costs and how far credits go on each plan, see OpenClaw API Costs Explained and How to Choose an OpenClaw Model Without Overpaying.