You don’t need to spend a penny to run OpenClaw. Here’s exactly how.
Most people assume running OpenClaw (formerly ClawdBot, briefly MoltBot) requires a paid API subscription. It doesn’t. Between OpenRouter’s free tier, provider free plans, and OAuth-based access, you can run a fully functional OpenClaw agent — across Telegram, Discord, WhatsApp, or whatever your platform of choice is — without spending anything.
The catch? Free models have tradeoffs. This guide covers what’s actually available right now, how to set each option up, and where the free tier stops being good enough.
The Free Model Landscape for OpenClaw
As of February 2026, there are four practical routes to running OpenClaw for free:
- Kimi K2.5 — free on OpenRouter (and via Moonshot directly)
- Pony Alpha — free on OpenRouter
- Minimax — 7-day free trial with generous limits
- ChatGPT Codex — free via OpenAI OAuth (requires OpenAI subscription for Codex, but free for Plus/Pro users)
Plus the perennial options: Google Gemini’s free tier, Groq’s free API, and the OpenRouter free model router that randomly selects from all available free models.
Let’s break each one down.
1. Kimi K2.5 — The Current King of Free
Kimi K2.5 from Moonshot AI is currently the #1 most-used model on OpenRouter for OpenClaw users. That’s not hype — it’s usage data. Developers chose it because it delivers reasoning quality close to Claude Sonnet 4.5 at zero cost.
What you get: Full tool-calling support, 128K context window, strong coding and reasoning. Moonshot has announced permanent free access on their platform.
Limits: Rate-limited on OpenRouter’s free tier (roughly 20 requests/minute, resets daily). Via Moonshot’s own API, limits are more generous but require a separate account.
Setup (OpenRouter):
openclaw config set agents.defaults.model.primary "openrouter/moonshotai/kimi-k2.5"
openclaw gateway restart
Or in openclaw.json:
{
"agents": {
"defaults": {
"model": {
"primary": "openrouter/moonshotai/kimi-k2.5"
}
}
}
}
Best for: General-purpose agent work, coding tasks, research, content drafting. This is the one to start with.
Quality: ★★★★☆ — Genuinely competitive with mid-tier paid models. Occasional weaknesses on very complex multi-step reasoning.
2. Pony Alpha — Surprisingly Capable
Pony Alpha is a lesser-known free model available on OpenRouter. It punches above its weight for a free option, with decent tool-calling support and reasonable context handling.
Limits: Free tier on OpenRouter with standard rate limits.
Setup:
openclaw config set agents.defaults.model.primary "openrouter/pony-alpha"
openclaw gateway restart
Best for: Lighter tasks — quick lookups, simple automations, heartbeat duties. Works well as a fallback model in a tiered setup.
Quality: ★★★☆☆ — Good for simple tasks. Starts to struggle with complex reasoning chains or long multi-turn conversations.
3. Minimax — Premium Feel, Free Trial
Minimax offers a 7-day free trial with access to their flagship model. The community consensus is that it’s better for writing and conversational tasks than for coding — the “vibes” model, as the OpenClaw docs put it.
Limits: 7-day trial period. After that, you’ll need to pay or switch.
Setup:
openclaw config set agents.defaults.model.primary "openrouter/minimax/minimax-01"
openclaw gateway restart
Best for: Content creation, conversational agents, creative tasks. If you’re using OpenClaw primarily for writing, try Minimax first.
Quality: ★★★★☆ — Excellent writing quality. Weaker on structured tool use and code generation compared to Kimi K2.5.
4. ChatGPT Codex via OAuth
If you already have an OpenAI Plus or Pro subscription, you can connect OpenClaw to ChatGPT’s Codex model via OAuth — no separate API key needed. Technically this isn’t “free” if you’re paying for ChatGPT, but if you’re already a subscriber, it costs nothing extra.
Setup:
openclaw onboard
# Follow the OAuth flow in your browser
Best for: Users who already pay for ChatGPT and want to use OpenClaw without additional API costs.
Quality: ★★★★☆ — Solid general-purpose model with good tool calling.
5. Bonus: Google Gemini Free Tier
Often overlooked, Google’s Gemini API has one of the most generous free tiers in the industry: 1,500 requests/day and 1 million tokens/minute on Gemini Flash.
Setup:
openclaw config set agents.defaults.model.primary "google/gemini-2.5-flash"
# Set your Gemini API key via openclaw onboard or:
openclaw config set models.providers.google.apiKey "YOUR_KEY"
openclaw gateway restart
Best for: High-volume usage. If you’re hitting rate limits on OpenRouter free models, Gemini Flash’s free tier is absurdly generous.
Quality: ★★★★☆ — Fast, capable, excellent for most daily tasks. Not as strong as frontier models on complex reasoning.
6. OpenRouter Free Model Router
Don’t want to pick? OpenRouter offers openrouter/free — a meta-model that randomly routes your requests to whichever free models are currently available.
openclaw config set agents.defaults.model.primary "openrouter/openrouter/free"
openclaw config unset agents.defaults.models
openclaw gateway restart
Quality: ★★½☆☆ — Inconsistent by design. Fine for testing, not ideal for production use since you can’t predict which model will respond.
Comparison Table
| Model | Provider | Free Tier Limits | Quality | Speed | Best For |
|---|---|---|---|---|---|
| Kimi K2.5 | OpenRouter / Moonshot | ~20 req/min, daily reset | ★★★★ | Fast | General agent work, coding |
| Pony Alpha | OpenRouter | Standard free limits | ★★★ | Fast | Simple tasks, fallback |
| Minimax | OpenRouter | 7-day trial | ★★★★ | Medium | Writing, creative tasks |
| Codex (OAuth) | OpenAI | Tied to ChatGPT sub | ★★★★ | Medium | Existing ChatGPT users |
| Gemini Flash | 1,500 req/day, 1M tok/min | ★★★★ | Very fast | High-volume daily use | |
| Groq (Llama 3.1 70B) | Groq | 14,400 req/day | ★★★½ | Fastest | Speed-critical tasks |
| openrouter/free | OpenRouter | Varies | ★★½ | Varies | Experimentation |
The Honest Tradeoff: What Free Can and Can’t Do
Free models handle well:
- Heartbeat checks and background monitoring
- Simple lookups (calendar, weather, quick searches)
- Drafting messages and basic content
- Straightforward coding tasks (single-file edits, scripts)
- Sub-agent coordination tasks
- Running ClawdBot or MoltBot as a basic personal assistant
Where you’ll want a paid model:
- Complex multi-file refactoring
- Architectural decisions requiring deep reasoning
- Long, nuanced multi-turn conversations with high context dependence
- Tasks requiring consistent, predictable output quality
- Production-critical workflows where random model failures aren’t acceptable
The smart move? Use a tiered setup. Put Kimi K2.5 or Gemini Flash as your primary, and save Opus or GPT-5.2 as a fallback for when you need frontier-level reasoning:
{
"agents": {
"defaults": {
"model": {
"primary": "openrouter/moonshotai/kimi-k2.5",
"fallbacks": ["google/gemini-2.5-flash"]
},
"heartbeat": {
"model": "google/gemini-2.5-flash-lite"
}
}
}
}
This gives you a completely free OpenClaw setup with a fast, capable primary model and a generous-free-tier fallback.
Free vs. Cheap: Is $2/Month Worth It?
If you’re willing to spend even a small amount, your options expand dramatically. Here’s what a few dollars gets you:
| Budget | Setup | What Changes |
|---|---|---|
| $0/mo | Kimi K2.5 + Gemini Flash free tiers | Fully functional, rate-limited |
| $1-2/mo | DeepSeek V3.2 ($0.53/M tokens) as primary | No rate limits, consistent quality |
| $3-5/mo | DeepSeek R1 for reasoning + Flash for simple tasks | Near-frontier reasoning at budget price |
| $10-20/mo | Claude Sonnet 4.5 or GPT-5 as primary | Frontier quality, no compromises |
For most MoltBot and ClawdBot users doing personal assistant tasks, the free tier is genuinely enough. If you’re using OpenClaw for professional development work, even $2/month on DeepSeek removes every meaningful limitation.
Getting Started in 5 Minutes
- Install OpenClaw if you haven’t: follow docs.openclaw.ai
- Get an OpenRouter API key at openrouter.ai (free, no credit card)
- Run the setup:
openclaw config set agents.defaults.model.primary "openrouter/moonshotai/kimi-k2.5"
openclaw config set models.providers.openrouter.apiKey "YOUR_OPENROUTER_KEY"
openclaw gateway restart
- Scan available free models:
openclaw models scan- Start chatting. Switch models anytime with
/modelin your conversation.
That’s it. Zero dollars. A fully functional AI agent running on your machine, connected to your messaging platform of choice.
Bottom Line
The openclaw free api ecosystem is better than it’s ever been. Kimi K2.5 alone makes the free tier viable for serious daily use — there’s a reason it’s the most-used model on OpenRouter right now. Pair it with Gemini Flash as a fallback and you’ve got a setup that costs nothing and handles 90% of what most people need.
The remaining 10%? That’s what /model opus is for. But you might be surprised how rarely you need it.
Last updated: February 2026