Claude Max subscribers using OpenClaw now pay API rates. Here's the math.
On April 4, Anthropic stopped counting third-party tool usage against your Claude subscription. OpenClaw got cut first. Every other harness follows. If you're a heavy user, the cost jump can be 16x or more.
Photo by Nathan Dumlao on Unsplash
TL;DR
- Effective April 4, 2026, Claude Pro and Max subscribers can no longer use their subscription quota through third-party agent tools like OpenClaw.
- Those tools now draw from pay-as-you-go API billing: Sonnet 4.6 at $3/$15 per MTok, Opus 4.6 at $5/$25 per MTok.
- Heavy users who were running $100+ in daily API-equivalent compute under a $200/month plan face real cost jumps - in documented cases, $3,286/month vs. $200/month before.
- Anthropic is offering a one-time credit equal to your plan cost ($100 or $200), redeemable by April 17, 2026. Extra usage bundles available at up to 30% off during transition.
- Official Anthropic tools (Claude Code CLI, claude.ai, Claude Desktop, Claude Cowork, Cursor) still count against your subscription. Only third-party harnesses are affected.
What changed on April 4
At noon Pacific on April 4, Anthropic blocked Claude Pro and Max subscribers from routing usage through third-party agent harnesses under their subscription quota. The first tool cut off was OpenClaw - an open-source agentic coding framework with 247,000 GitHub stars and over 135,000 running instances. Anthropic said the restriction would extend to all third-party harnesses "in the coming weeks."
Before April 4: third-party tools used your subscription's included quota. The Max 20x plan gave you 20 times the base allowance. If you ran OpenClaw all day on Opus 4.6, that came out of your $200/month bucket.
After April 4: third-party tools draw from "extra usage," billed separately at standard API rates. Your subscription still covers official surfaces. But anything routed through OpenClaw - or any other third-party harness going forward - now generates API invoices on top of your subscription cost.
Who gets hit and who doesn't
| Tool | Status after Apr 4 |
|---|---|
| Claude Code CLI (official) | Covered by subscription - no change |
| claude.ai web interface | Covered by subscription - no change |
| Claude Desktop | Covered by subscription - no change |
| Claude Cowork | Covered by subscription - no change |
| Cursor (official Anthropic partner) | Covered by subscription - no change |
| OpenClaw | Now billed at API rates |
| All other third-party harnesses | API billing rollout in coming weeks |
Note: projects that proxied Claude Max subscriptions as API endpoints (used to run OpenCode, Aider, Crush, Cline, etc.) were already in a gray area and are covered by this policy as well.
What third-party usage costs at API rates
These are the rates that now apply to OpenClaw and future third-party tool usage:
| Model | Input / 1M tokens | Output / 1M tokens | Batch (50% off) |
|---|---|---|---|
| Claude Haiku 4.5 | $1.00 | $5.00 | $0.50 / $2.50 |
| Claude Sonnet 4.6 | $3.00 | $15.00 | $1.50 / $7.50 |
| Claude Opus 4.6 | $5.00 | $25.00 | $2.50 / $12.50 |
Prompt caching discounts available: Sonnet 4.6 cache reads at $0.30/MTok (a tenth of the base input rate). Extended context above 200K tokens: Sonnet 4.6 input doubles to $6/MTok, Opus 4.6 input doubles to $10/MTok. Rates from Anthropic pricing, retrieved April 8, 2026.
Third-party harnesses like OpenClaw also tend to bypass Anthropic's prompt cache optimization layer, meaning a session that official Claude Code would cache aggressively instead sends full context on every call. Boris Cherny, head of Claude Code at Anthropic, cited this directly: "Our systems are highly optimized for one kind of workload... subscriptions were never designed for the kind of continuous, automated demand these tools generate."
The cost math for different usage levels
In January 2026, German tech publication c't 3003 ran OpenClaw for a single day on Claude Opus 4.6 and measured the API-equivalent cost at $109.55. That usage was covered under a $200/month subscription at the time - roughly $6.67/day amortized. The same day of usage now costs $109. Monthly extrapolation: $3,286 vs. $200.
Here's the math across more realistic usage levels on Sonnet 4.6 with a typical 60/40 input/output split:
Prompt caching can reduce these numbers significantly if your harness supports it. Anthropic's own estimate for official Claude Code users is $6/day average, $12/day at the 90th percentile - but that baseline already reflects caching optimizations that third-party harnesses typically don't use. Use the cost calculator for a custom estimate.
Anthropic's explanation (and the part they didn't say)
Boris Cherny framed this as a capacity problem: "Anthropic's subscriptions weren't built for the usage patterns of these third-party tools. Capacity is a resource we manage thoughtfully." The underlying argument is that third-party harnesses send full context on every API call, bypassing Anthropic's caching infrastructure. The same amount of "useful work" costs Anthropic meaningfully more to serve via OpenClaw than via official Claude Code.
That argument has some technical merit. It also has a convenient side effect for Anthropic's revenue. Some developers on Hacker News were more direct about what they saw: "Anthropic oversold their server capacity, and this was their response, point blank and simple."
What Anthropic didn't say: the timing overlaps with a period where they added several OpenClaw-like capabilities to official Claude Code - multi-agent support, the March 24 computer control launch, improved tool use. The restriction arrived after official Claude Code became a more capable competitor to third-party harnesses, not before.
The OpenClaw situation
OpenClaw hit 100,000 GitHub stars in under 48 hours in late February - one of the fastest open-source growth rates on record for a developer tool. By the time Anthropic enforced the restriction, the project had 247,000 stars, 47,700 forks, and over 50 integrations across Claude, GPT-4o, Gemini, and DeepSeek. About 135,000 instances were running at any given time.
Peter Steinberger, the creator, had already joined OpenAI in February 2026. He learned about the restriction a week in advance and spent that time trying to negotiate a different outcome: "Both me and @davemorin tried to talk sense into Anthropic, best we managed was delaying this for a week."
OpenClaw works fine on the direct API and on models from other providers. Steinberger said OpenAI's infrastructure would support continued open-source development. The project isn't going away - it just no longer benefits from Anthropic's best models at subscription prices.
The transition offer (expires April 17)
The one-time credit softens the immediate hit. For moderate users whose monthly API spend might run $100-$200 anyway, it covers roughly a month of the new normal. For the heavy users - the ones running $109/day in compute - it's about 18 hours of Opus time.
What your options look like now
If you were using OpenClaw under a Claude Max subscription, the realistic paths forward:
Your subscription still covers this. If what you were doing with OpenClaw can be done through the official CLI, nothing changes except the interface. Multi-agent support, computer control, and tool use are all available in official Claude Code now.
OpenClaw supports GPT-5.4, Gemini 3, and DeepSeek V3.2 in addition to Claude. GPT-5.4 at $2.50/$15 per MTok or Gemini 3.1 Pro at $2/$12 per MTok are both viable alternatives depending on your task mix - see the pricing comparison for the full picture.
Cursor is an official Anthropic partner integration and is not affected. It has access to Claude Sonnet 4.6 and Opus 4.6 through its own subscription model ($20/month Pro, ~$200/month Ultra). If you were using OpenClaw primarily as an editor interface, Cursor handles most of the same agentic workflows.
The Batch API cuts all rates by 50% (Sonnet 4.6 drops to $1.50/$7.50). Prompt caching cuts repeated context reads by 90%. For async workloads, those two together can bring actual spend well below headline prices.
| Option | Monthly cost | Model access |
|---|---|---|
| Claude Max 20x (official tools only) | $200/mo | Sonnet 4.6, Opus 4.6 via official surfaces |
| Claude Max 20x + API for third-party | $200 + API usage | Same, plus any harness at pay-per-token rates |
| Cursor Ultra | ~$200/mo | Claude Sonnet 4.6, Opus 4.6, GPT-5.4, Gemini 3 Pro, others |
| GitHub Copilot Pro+ | $39/mo | Claude Opus 4, OpenAI o3, GPT-5.4 |
| Direct API (no subscription) | Pay per use | All Claude models at listed rates |
Compare all model pricing on our pricing page.
Sources
- TechCrunch: Anthropic says Claude Code subscribers will need to pay extra for OpenClaw support (April 4, 2026)
- The Next Web: Anthropic OpenClaw Claude subscription ban cost
- InfoWorld: Anthropic cuts OpenClaw access from Claude subscriptions
- VentureBeat: Anthropic cuts off Claude subscriptions for OpenClaw
- Hacker News discussion thread
- Anthropic: Manage extra usage for paid Claude plans
- Anthropic pricing page (retrieved April 8, 2026)