How to Set Up Claude Opus 4.6 on OpenClaw — Step by Step

Why You Should Do This Right Now

On February 5, 2026, Anthropic released Claude Opus 4.6. It's a major upgrade over Opus 4.5:

• ●1 million token context window (5x larger than Opus 4.5's 200K)
• ●128K token output — generate entire documents, not just snippets
• ●Agent teams — multiple AI agents working in parallel on the same task
• ●Adaptive thinking — the model adjusts reasoning depth based on complexity
• ●Context compaction — auto-summarizes old context so long sessions never crash

It also set new records on Terminal-Bench 2.0 (agentic coding), GDPval-AA (enterprise tasks), and BrowseComp (web research). And pricing is unchanged: $5/$25 per million tokens.

If you're running OpenClaw, here's how to switch to it in under 5 minutes.

Prerequisites

• ●OpenClaw installed (npm install -g openclaw@latest)
• ●Anthropic API key configured

If you haven't set up OpenClaw yet, start with the official docs.

Step 1: Edit the Config File

Open your OpenClaw config at ~/.openclaw/openclaw.json and add or update these two sections.

Define Opus 4.6 in models

"models": {
  "mode": "merge",
  "providers": {
    "anthropic": {
      "baseUrl": "https://api.anthropic.com",
      "api": "anthropic-messages",
      "models": [
        {
          "id": "claude-opus-4-6",
          "name": "Claude Opus 4.6",
          "reasoning": true,
          "input": ["text", "image"],
          "contextWindow": 1000000,
          "maxTokens": 128000
        }
      ]
    }
  }
}

What matters here:

• ●mode: "merge" — adds to OpenClaw's built-in model catalog without replacing anything
• ●reasoning: true — enables Opus 4.6's extended thinking mode
• ●contextWindow: 1000000 — the full 1M token context
• ●maxTokens: 128000 — 128K output capability

Set Opus 4.6 as default in agents

"agents": {
  "defaults": {
    "model": {
      "primary": "anthropic/claude-opus-4-6",
      "fallbacks": [
        "anthropic/claude-opus-4-5"
      ]
    },
    "contextTokens": 1000000
  }
}

This sets Opus 4.6 as the default model for all sessions, with Opus 4.5 as a fallback. The contextTokens value tells the agent it can use the full 1M context window.

Step 2: Restart the Gateway

After saving the config, restart OpenClaw to reload it:

openclaw gateway restart

Step 3: Start a New Session

Existing sessions keep their old model config. You need a fresh session:

/new

Or /reset. The new model kicks in from the next session onward.

Step 4: Verify

Check that everything is wired up correctly:

openclaw models status

You should see anthropic/claude-opus-4-6 listed as the primary model.

From inside a chat session, you can also run:

/model status

If it shows Opus 4.6, you're good.

What's Actually Different Day-to-Day

Beyond the spec sheet numbers, here's what you'll notice in practice:

• ●Longer sessions without drift — the model doesn't "forget" what you discussed earlier in the conversation. The 1M context window eliminates context rot.
• ●Better at catching its own mistakes — Opus 4.6 reviews its own code and reasoning more carefully. It'll spot bugs it introduced and fix them without you pointing it out.
• ●More autonomous on complex tasks — it breaks big tasks into steps, works through them, and recovers from errors along the way. Less hand-holding needed.
• ●Faster iteration on documents — spreadsheets, presentations, and reports come out closer to production-ready on the first attempt.

Quick Reference

| Step | Command | |------|---------| | Install/update OpenClaw | npm install -g openclaw@latest | | Edit config | ~/.openclaw/openclaw.json | | Restart gateway | openclaw gateway restart | | New session | /new or /reset | | Verify model | openclaw models status |

References

• ●OpenClaw Docs — Models
• ●OpenClaw Docs — Model Providers
• ●OpenClaw GitHub
• ●Anthropic — Claude Opus 4.6