Access Claude Opus, GPT, and Gemini in One $10/Month Subscription — No API Keys, No Per-Token Bills

By James Tipton & Chief Wizard —  |  Category: AI Engineering

The Problem: Anthropic Blocked Claude in OpenClaw

If you've tried to use your Claude Pro subscription with OpenClaw, you've hit the wall. On April 4, 2026, Anthropic officially cut off subscription-based OAuth access for all third-party tools. Your $20/month Claude Pro plan works great on Claude.ai and Claude Code — but it won't power your OpenClaw agents.

The workaround most people try is an Anthropic API key. That works, but API billing for an always-on AI agent adds up fast. Heavy use runs $50-200/month. You lose the flat-rate predictability that made subscriptions attractive in the first place.

There's a better path.

The Solution: GitHub Copilot as Your Model Provider

GitHub Copilot is Microsoft's AI coding assistant at $10/month. What most people don't realize is that Copilot's backend gives you access to multiple frontier AI models — including Claude Opus, GPT, and Gemini — through a single subscription.

OpenClaw has built-in support for GitHub Copilot as a model provider. One login command, and you have access to all of Copilot's models inside OpenClaw via OAuth. No API keys to manage, no per-token billing, no Anthropic restrictions.

Why Claude Opus 4.5 Is Worth It

Before we get into setup, let's talk about why you'd want Claude in your OpenClaw agent in the first place.

The real power: you can assign different models to different agents. CFO running deep financial analysis? Claude Opus. CTO doing code review? GPT Codex. CMO generating social content? Claude Sonnet for speed. All from one $10/month subscription.

Cost Comparison

GitHub Copilot is the best value in the ecosystem right now — multiple frontier models, flat rate, officially supported.

1 Sign Up for GitHub Copilot

Go to github.com/features/copilot and sign up for the Individual plan at $10/month. You'll need a GitHub account — sign up with your business email if you have one.

2 SSH Into Your OpenClaw Server

If you're running OpenClaw on a local VM or VPS, SSH in as normal. The next command requires an interactive terminal — it can't be run inside a script or automation.

ssh user@your-server-ip

3 Run the Copilot Login Command

This is the entire setup. One command:

openclaw models auth login-github-copilot

OpenClaw will display a URL and a one-time code. Open the URL in your browser, enter the code, and authorize with your GitHub account. Keep the terminal open until it confirms completion.

4 Set Claude Opus as Your Default Model

Now tell OpenClaw to use Claude Opus as the primary model:

openclaw config set agents.defaults.model.primary "github-copilot/claude-opus-4.5"

Set GPT as a fallback in case Copilot hits a rate limit:

openclaw config set agents.defaults.model.fallbacks '["openai-codex/gpt-5.4"]' --strict-json

Update the allowed models list to match:

openclaw config set agents.defaults.models '{"github-copilot/claude-opus-4.5":{},"openai-codex/gpt-5.4":{}}' --strict-json

5 Restart the Gateway

openclaw gateway restart

Verify the model loaded correctly:

openclaw models status

You should see:

Default: github-copilot/claude-opus-4.5

And critically — no warmup error in the logs:

tail -20 /tmp/openclaw/openclaw-$(date +%Y-%m-%d).log | grep -i "warmup\|model"

If you see agent model: github-copilot/claude-opus-4.5 with no warmup failure — you're done.

6 Test Your Agent

Message your bot and ask it to check its own model:

"Use your session_status tool and tell me exactly what model you are running"

It should respond with github-copilot/claude-opus-4.5. If it's still reporting GPT, clear the session cache and restart:

rm -rf ~/.openclaw/agents/main/sessions/
openclaw gateway restart

That's Claude. You'll know it when you feel it.

Available Models via GitHub Copilot

Multi-Agent Setup: Different Models for Different Roles

One of the most powerful things about this setup is assigning different models to different agents. Here's how to configure a CFO agent that uses Claude Opus for deep financial analysis while your main assistant uses a faster model:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "github-copilot/claude-opus-4.5"
      }
    },
    "list": [
      { "id": "main" },
      {
        "id": "cfo",
        "model": "github-copilot/claude-opus-4.5"
      },
      {
        "id": "coder",
        "model": "github-copilot/gpt-5.2-codex"
      }
    ]
  }
}

Each agent gets the brain best suited for its role. All from one $10/month subscription.

Troubleshooting

Agent still reports GPT after setup

Clear the session cache and restart:

rm -rf ~/.openclaw/agents/main/sessions/
openclaw gateway restart

Warmup error in logs

Check the model string — it must use dots not dashes:

openclaw config set agents.defaults.model.primary "github-copilot/claude-opus-4.5"

Then verify the allowed models list matches exactly:

cat ~/.openclaw/openclaw.json | grep -A5 "models"

Authentication failed

Re-run the login flow:

openclaw models auth login-github-copilot

Model not available

Check what models your plan includes:

openclaw models list --provider github-copilot

Power Your AI Business

Two tools we actually use — one for infrastructure, one for the builder behind the bot.

Keywords: OpenClaw Claude, GitHub Copilot OpenClaw, Claude Opus OpenClaw, use Claude without API key, OpenClaw GitHub Copilot setup, Claude Opus 4.5 OpenClaw, free Claude API alternative, OpenClaw model provider, Anthropic restriction workaround, Claude in Telegram bot