Anthropic did not block third-party tools from Claude on April 4, 2026. That happened months ago. What changed today is the price.
Starting at noon Pacific, Claude Pro and Max subscriptions no longer cover usage routed through third-party tools. Subscribers who had been using OpenClaw, OpenCode, or any external tool with their subscription credentials must now pay through a separate “extra usage” billing tier (pay-as-you-go, metered per token) or authenticate with a standard API key. Anthropic is compensating every Pro and Max subscriber with a one-time credit equal to one month of subscription cost, redeemable by April 17, plus up to 30% off pre-purchased extra usage bundles.
The distinction matters. Third-party tools were already forbidden from accessing Claude subscriptions. Anthropic began enforcing this in January 2026, when engineer Thariq Shihipar deployed server-side blocks against tools spoofing the Claude Code authentication flow. By February 20, the company had revised its legal terms to explicitly restrict OAuth tokens to Claude Code and Claude.ai. By March, OpenCode had stripped all Claude subscription authentication from its codebase after receiving legal demands. The blocking is old news.
The new news is economic. Anthropic formalized the pricing tier that separates first-party and third-party compute. If you use Claude through Anthropic’s own products (Claude.ai, Claude Code, Claude Cowork, the desktop app), your subscription covers it. If you use Claude through anything else, you pay per token.
Why the Price Difference Is Structural
The pricing split is not arbitrary. It reflects a real cost asymmetry between first-party and third-party usage, driven by prompt cache optimization.
Claude Code is engineered to maximize cache hit rates. When a developer works in Claude Code, the tool reuses previously processed context across requests. A cache hit on Opus 4.6 costs $0.50 per million input tokens. An uncached request costs $5.00. That 90% reduction is what makes flat-rate subscriptions economically viable for Anthropic’s own tools. The effective cost of serving a Claude Code session is a fraction of the nominal per-token rate because most context is already cached.
Third-party tools construct their own prompts and manage their own context windows. Their requests rarely align with Anthropic’s caching infrastructure. Every request is more likely to be a full-price cache miss. The cost gap between a Claude Code session and an equivalent OpenClaw session producing the same output can be 5x to 25x, according to industry estimates.
The subscription credit and the extra usage tier are Anthropic’s way of saying: we will no longer absorb the cost differential, but we will give you a path to keep using external tools at metered rates, and we will compensate you for the transition.
The Wider Pattern
Google enforced the same pricing split on Gemini CLI in March 2026. Accounts routing third-party traffic through Gemini CLI’s OAuth flow were flagged, some banned, and free-tier users lost access to Pro models entirely. The same structural economics apply: flat-rate subscriptions priced for human-speed usage cannot sustain autonomous agent loops that run at machine speed.
OpenAI took the opposite position. OpenClaw’s documentation now steers users toward OpenAI as the default path. Whether OpenAI sustains this as compute pressure mounts is an open question.
The pricing tier is only one dimension of the vendor lock-in pattern. The harder economic question is what gets optimized at the tool layer. Our analysis of the edit tool benchmark that improved 15 LLMs without touching a single weight shows that the largest performance gains in agentic coding now live in open-source infrastructure that first-party vendors have no incentive to build.
What to Do
If you use Claude exclusively through Claude.ai, Claude Code, or the desktop app: nothing changed. Your subscription covers everything.
If you use Claude through third-party tools: you now pay per token via extra usage or API key. Instrument your token consumption before enabling metered billing. With prompt caching (90% input cost reduction) and batch processing (50% discount), the actual cost increase with proper engineering is 1.5x to 3x, not the 5x to 25x sticker shock that assumes worst-case unoptimized usage.
Claim the credit before April 17. Every Pro and Max subscriber qualifies regardless of whether you used third-party tools.
Evaluate whether your workflows can migrate to Claude Code. It remains subscription-covered, benefits from 90% cache cost reduction, and supports team-shared configurations through the .claude/ protocol system.