Anthropic announced on April 4, 2026 that Claude Pro and Max subscription credits no longer cover usage in third-party tools — including OpenClaw, Claude Code, and other API-connected apps. Going forward, only two billing methods work with OpenClaw: API Key (pay-as-you-go) or Extra Usage credits (prepaid blocks). Anthropic is offering a one-time refund credit for affected users and up to 30% off Extra Usage pre-purchases through April 17.
What This Means for OpenClaw Users
If you were running OpenClaw on a Claude Pro ($20/mo) or Max ($100/mo) subscription and using those credits for your agent — that billing path is gone. Your subscription still works for claude.ai directly, but any API calls made by OpenClaw now bill separately against your API key.
For most users this means one of three things:
- Your OpenClaw stopped working — if you had no API key configured, calls are now failing silently or returning auth errors
- Your costs just jumped — if you had an API key configured, it's now absorbing all OpenClaw usage that was previously covered by your subscription
- Nothing changed — if you were already running purely on API keys with no subscription credits applied to OpenClaw
Your Options Going Forward
Act before April 17: Anthropic is offering a one-time refund credit for users affected by this change, plus up to 30% off Extra Usage pre-purchases. Both offers expire April 17. Log into your Anthropic account to claim.
How to Set Up Spend Limits (Do This Today)
If you're now running on API keys only, the single most important thing you can do is set a hard spend cap before your agent racks up unexpected charges. Here's how:
- Go to console.anthropic.com → Settings → Billing
- Under Usage limits, set a monthly spend limit (start with $50 if you're unsure)
- Set an email alert at 80% of your limit — you'll get notified before hitting the cap
- In your openclaw.json, set
"model": "claude-sonnet-4-6"instead of Opus for routine tasks — Sonnet costs ~5x less per token
Without a spend cap: A heartbeat loop running every 30 minutes on Opus can generate $100–300+ in a single day if something goes wrong. This is not hypothetical — we've seen it happen to users who assumed their subscription would absorb overages.
The Local Model Strategy (Best Long-Term Move)
The cleanest response to this change is a hybrid routing strategy: run free local models for 80% of tasks, use Claude API only when you actually need its reasoning capability.
A practical split for most business operators:
- Local (free): Heartbeat checks, memory reads, scheduling, simple lookups, draft generation
- Claude API: Complex analysis, SOUL.md interpretation, multi-step planning, anything client-facing
On a Mac Mini M4 or mid-range Linux box, Qwen 3.6B or Llama 3.3 8B run fast enough for routine tasks. Setup takes about 20 minutes with Ollama — we have a full guide at /blog/openclaw-ollama-local-models.
Most users who implement this cut their monthly API spend from $60–100 down to $10–20 for the same workload.
Quick Cost Reference (Post-Change)
As of April 2026, approximate Claude API costs per million tokens:
- Claude Opus 4.6: ~$15 input / $75 output — use for complex reasoning only
- Claude Sonnet 4.6: ~$3 input / $15 output — good default for most tasks
- Claude Haiku: ~$0.25 input / $1.25 output — fast and cheap for simple tasks
- Local (Ollama): $0 — requires hardware, latency is higher
A well-configured OpenClaw running Sonnet as default with local for heartbeats costs most users $15–40/month in API spend. Opus-only setups can run $150–400/month depending on usage.
If you need help reconfiguring your setup after this change — model routing, spend limits, Ollama integration — book a free call. We can usually sort it out in under an hour.