OpenClaw Model Routing Guide: Which AI Model for Which Task

Stop using the most expensive model for everything. This guide gives you copy-paste configurations that cut your OpenClaw bill by 50–80% without sacrificing quality.

The Problem: One Model Does Not Fit All

OpenClaw supports dozens of AI models via OpenRouter, Anthropic, OpenAI, Google, and more. That's incredibly powerful — but it creates a paradox of choice. Most new users pick the biggest, smartest model (Claude Opus 4, GPT-5.2) and use it for everything: heartbeats, quick questions, cron jobs, sub-agents.

The result? $100–200+/month bills when $20–50 would get you the same experience for 95% of tasks.

This guide gives you a practical model routing strategy: which model for which task, with actual OpenClaw openclaw.json configurations you can copy-paste right now.

Model Tiers & Pricing (February 2026)

Here's how the current model landscape breaks down into cost tiers. Prices are per million tokens (input/output).^[1][2][3]

🏆 Tier 1 — Flagship (Use Sparingly)

⚡ Tier 2 — Workhorse (Daily Driver)

💰 Tier 3 — Budget (Bulk/Simple Tasks)

🆓 Tier 4 — Free/Local

Task-to-Model Mapping: The Cheat Sheet

This is the core of the guide. For every common OpenClaw task, here's which model to use and why:

OpenClaw Configuration Examples

Here are three ready-to-use configurations for different budgets. OpenClaw uses a ~/.openclaw/openclaw.json file in JSON5 format.^[4]

💵 Budget Setup ($5–15/month)

Use cheap models everywhere, with Sonnet only for main chat sessions.

{
  agents: {
    defaults: {
      model: {
        primary: "anthropic/claude-haiku-4-5",
        fallbacks: ["openai/gpt-5-mini"],
      },
      models: {
        "anthropic/claude-haiku-4-5": { alias: "haiku" },
        "anthropic/claude-sonnet-4-6": { alias: "sonnet" },
        "openai/gpt-5-mini": { alias: "gpt-mini" },
      },
      heartbeat: {
        every: "30m",
        model: "anthropic/claude-haiku-4-5",
      },
    },
    list: [
      {
        id: "main",
        default: true,
        // Override to Sonnet for your main interactive session
        model: "anthropic/claude-sonnet-4-6",
      },
    ],
  },
  cron: { enabled: true },
}

With this setup, heartbeats and cron jobs use Haiku (~$1/MTok input), sub-agents default to Haiku, and only your direct conversations use Sonnet. If Haiku is down, it falls back to GPT-5 mini.

⚖️ Balanced Setup ($20–50/month)

Sonnet for most tasks, Haiku for background work. The sweet spot for most users.

{
  agents: {
    defaults: {
      model: {
        primary: "anthropic/claude-sonnet-4-6",
        fallbacks: ["openai/gpt-5.2"],
      },
      models: {
        "anthropic/claude-opus-4-6": { alias: "opus" },
        "anthropic/claude-sonnet-4-6": { alias: "sonnet" },
        "anthropic/claude-haiku-4-5": { alias: "haiku" },
        "openai/gpt-5.2": { alias: "gpt" },
        "openai/gpt-5-mini": { alias: "gpt-mini" },
      },
      heartbeat: {
        every: "30m",
        model: "anthropic/claude-haiku-4-5",
      },
    },
  },
  cron: { enabled: true },
}

Everything defaults to Sonnet, but heartbeats run on Haiku. Use /model opus in chat when you need Opus for a complex task, then switch back. Cron jobs inherit the default model — override with per-job model settings for savings.

🚀 Power Setup ($50–100/month)

Sonnet as default with Opus readily available. For power users who need the best for complex tasks.

{
  agents: {
    defaults: {
      model: {
        primary: "anthropic/claude-sonnet-4-6",
        fallbacks: ["anthropic/claude-opus-4-6", "openai/gpt-5.2"],
      },
      models: {
        "anthropic/claude-opus-4-6": { alias: "opus" },
        "anthropic/claude-sonnet-4-6": { alias: "sonnet" },
        "anthropic/claude-haiku-4-5": { alias: "haiku" },
        "openai/gpt-5.2": { alias: "gpt" },
        "openai/gpt-5-mini": { alias: "gpt-mini" },
        "google/gemini-3-flash-preview": { alias: "flash" },
      },
      heartbeat: {
        every: "30m",
        model: "anthropic/claude-haiku-4-5",
      },
      maxConcurrent: 3,
    },
  },
  cron: { enabled: true },
}

If Sonnet goes down, it falls back to Opus, then GPT-5.2. Use /model opus for complex reasoning tasks. Set maxConcurrent: 3 for parallel sub-agents.

Monthly Cost Estimates

Estimated monthly costs based on typical usage patterns. Assumes ~2K tokens per message (input+output combined).^[5]

Light Use (20 messages/day)

Medium Use (50 messages/day + 5 cron jobs)

Heavy Use (100 messages/day + 10 cron jobs + sub-agents)

Model Routing with OpenRouter

OpenRouter is a unified API gateway that gives you access to 300+ models from all major providers with a single API key. It's the easiest way to use multiple models in OpenClaw.^[6]

Why OpenRouter?

• One API key, all models — No need for separate Anthropic, OpenAI, and Google accounts
• Automatic fallbacks — If one provider is down, OpenRouter routes to another
• Transparent pricing — Pay-as-you-go with a small markup over direct API pricing
• Free tier models — Some models have free tiers for testing

Using OpenRouter Models in OpenClaw

Prefix model IDs with openrouter/ in your config:

{
  agents: {
    defaults: {
      model: {
        primary: "openrouter/anthropic/claude-sonnet-4-6",
        fallbacks: ["openrouter/openai/gpt-5.2"],
      },
    },
  },
}

Per-Session Model Switching

In any OpenClaw chat, use the /model command to switch models on the fly:

• /model opus — Switch to Opus for a complex task
• /model haiku — Drop to Haiku for simple questions
• /model gpt — Try GPT for a second opinion

The model catalog in agents.defaults.models defines what's available. Aliases make switching fast.

What the Community Recommends

Based on Hacker News discussions, Reddit threads, and real-world OpenClaw deployments:^[7][8][9]

Consensus Picks

• Claude Sonnet is the overwhelming favorite for daily driver use — "best quality-to-cost ratio"
• Prompt caching is the #1 cost saver — reduces input token costs by up to 90% on repeated context
• Session resets (/clear) are the easiest win most people miss
• GPT-5 mini is a "value champion" — matches GPT-4o quality at a fraction of the cost^[10]
• Haiku for heartbeats — universal recommendation, no one uses flagship models for background checks
• DeepSeek models offer incredible value for coding tasks^[11]

Common Mistakes

• Using Opus/GPT-5.2 for everything — the biggest cost trap
• Not setting heartbeat model separately — burns flagship tokens on simple checks
• Letting conversations run forever without session resets
• Not using /model to switch models mid-conversation when the task changes
• Running 10+ cron jobs on the default (expensive) model

Hacker News Perspective

A popular HN thread "Ask HN: OpenClaw users, what is your token spend?" revealed most power users spend $20–60/month with smart routing, while those using a single flagship model everywhere reported $100–200/month for similar usage patterns.^[7]

Getting Started Checklist

Follow these steps to optimize your OpenClaw model routing today:

1. Sign up for OpenRouter — Get one API key at openrouter.ai for access to all models
2. Copy the Balanced Setup config — Paste it into ~/.openclaw/openclaw.json
3. Set Sonnet as your default — It's the best quality-to-cost ratio for daily use
4. Set Haiku for heartbeats — In agents.defaults.heartbeat.model
5. Configure model aliases — So you can /model opus quickly when needed
6. Enable session resets — Set session.reset.mode: "daily" or use /clear regularly
7. Monitor costs for 1 week — Check your OpenRouter dashboard or provider billing
8. Adjust based on actual usage — If you never use Opus, remove it. If Haiku isn't enough for cron jobs, upgrade selectively