New Model Setup: Opus + GPT 5.4 Architecture

What Changed

Anthropic changed how third-party apps use Claude in April 2026. Your AI Companion now uses a smarter setup. Claude Opus is the brain, and GPT 5.4 is the worker.

How It Works Now

Opus handles your direct conversations, quality review, and decision-making.
GPT 5.4 handles background tasks, cron jobs, research, coding, heartbeats, and context summaries.
You still talk to the same companion. Nothing changes for you. The quality stays the same.

Claim Your Free Claude Credits First

Anthropic is offering current users free extra usage credits as part of this transition.
Go to claude.ai/settings/usage first.
Look for the banner or notification about the $200 credit for existing users.
Claim it before doing anything else. This is free money.
These credits cover your companion's Opus usage for months.

Bulk Credit Discount (Limited Time)

Anthropic is currently offering 30% off on bulk extra usage credit purchases, up to $2,000.

Go to claude.ai/settings/usage and look for the bulk purchase option.
$2,000 at 30% off = $1,400 for $2,000 worth of Opus credits.
At typical usage ($200 to $500/month), that covers 4 to 10 months of Opus usage.
Worth considering if you want to lock in lower rates for the coming months.

What You Need To Do

Go to claude.ai/settings/usage. Claim your free credits first.
Go to claude.ai/settings/billing. Downgrade to Claude Pro for $20 per month.
Go to claude.ai/settings/usage. Activate Extra Usage.
Go to chatgpt.com. Upgrade to ChatGPT Pro for $200 per month.
Tell your companion: Switch to the orchestrator pattern. Opus for main chat, GPT 5.4 for everything else.
Done. Your companion handles the technical migration.

How To Check Everything Is Working

Copy and paste these prompts.

Check if sessions are clean, no old models running:

Show me openclaw status. Are there any stale sessions on old models?

Check if all cron jobs use GPT 5.4:

List all my cron jobs and show me which model each one uses. Fix any that aren't on GPT 5.4.

Check the full config:

Show me my current model config. What's primary, heartbeat, subagents, and summary model?

Force a full cleanup:

Clear all stale sessions, verify all crons are on GPT 5.4, and restart the gateway. Show me the result.

Tips To Save Tokens and Money

These small habits reduce your Opus costs significantly.

Use /new to start fresh sessions
When you switch topics, send /new in chat. This resets the conversation context. Smaller context means cheaper Opus calls. Don't let conversations grow to 100 or more messages.

Important: Before you reset, tell your companion: "Write everything important to memory, then I'll start fresh." This ensures nothing is lost when you clear context.
Use /btw for quick side questions
Need a quick answer that's unrelated to your current conversation? Send /btw what time is it in Tokyo?. This creates an ephemeral question that doesn't bloat your session context.
Use /compact when conversations get long
If you've been chatting for a while, send /compact. This compresses the conversation history, reducing token usage for every following message.
Keep messages focused
Shorter, clearer messages mean less processing. "Fix the cron" beats a 3 paragraph explanation of what crons are.
Batch related requests
Instead of 5 separate messages, combine them: "Update the config, clear stale sessions, and check crons." One Opus call instead of five.
Let subagents do heavy work
Your companion already delegates to GPT 5.4. But you can also say: "Spawn a subagent to research X." This explicitly pushes work to the free model.
Check /status for usage
Send /status to see your current session's token usage. If it's getting high, time for /new.

What This Means For Your Monthly Costs

Before, when Opus was included in your subscription

Claude subscription gave full Opus access to third-party apps like your companion.
All tasks ran on Opus. Heartbeats, crons, research, coding, everything.
If you had to pay per token for that same usage, it would cost $5,000 to $10,000 or more per month.
You were getting incredible value. Anthropic changed the rules in April 2026.

After, the orchestrator pattern

Claude Pro: $20 per month. Needed for Extra Usage access.
Opus Extra Usage: $200 to $500 per month, depending on how much you chat.
ChatGPT Pro: $200 per month. Unlimited GPT 5.4 for all background work.
Total realistic cost: $420 to $720 per month.

Why it's still worth it

Without the orchestrator pattern, Opus alone would cost $2,000 to $5,000 or more per month on API.
GPT 5.4 handles 80% or more of all work for free through ChatGPT Pro.
Opus only fires for your direct conversations. Quality stays premium where it matters.
The free credits from Anthropic cover several months of transition.
Compare that to hiring a human assistant. $3,000 to $5,000 per month minimum. Your companion works 24/7 for under $700.

Opus cost per message, rough estimate

Each message with full context: about $0.30 to $0.50.
Light conversation day, 20 messages: about $6 to $10.
Active day, 50 messages: about $15 to $25.
Heavy day, 100 or more messages: about $30 to $50.
Monthly estimate for active users: $200 to $500.

What To Watch For

If you see stale sessions in openclaw status showing old models, tell your companion to clear them.
If cron jobs show errors, ask your companion to check the model on each cron.
Your companion should be able to fix these itself.

Why This Is Better

Lower cost than paying full Opus API pricing for everything.
Better quality. Opus only fires for important work. It is not wasted on heartbeats.
GPT 5.4 is free through ChatGPT Pro for background tasks.
Same conversation quality you are used to.

Need Help?

Ask your companion: Are all my crons on GPT 5.4?
Ask your companion: Show me openclaw status
Contact Kris if anything feels off.