Anthropic Clarifies CLI Policy After User Backlash Over Usage Limits

TL;DR

Anthropic has officially clarified that using command-line interface (CLI) tools built on its API, such as OpenClaw, is permitted following user documentation updates. This policy clarification arrives amidst a period of significant user unrest over perceived performance degradation and aggressive usage limits in Claude Opus 4.6. The company has since released Claude Opus 4.7, focusing on advanced software engineering and addressing some user concerns with features like a new "xhigh" effort level and "task budgets" for better control over long-running tasks.

CLI Access Confirmed Amid Usage Limit Controversy

Developers can breathe a sigh of relief. Anthropic's documentation for OpenClaw, a third-party CLI tool, explicitly outlines configuration for using the Anthropic API key, effectively sanctioning this method of access. The setup process involves using the openclaw onboard command with an API key, with configuration snippets provided for models like Claude Opus 4.6. This official guidance resolves ambiguity and empowers developers to integrate Claude into automated and scripted workflows outside the standard web interface.

The confirmation comes at a critical juncture. For weeks, users across platforms like GitHub and Reddit have reported hitting usage limits far more quickly than expected, with some Max subscribers exhausting their 5-hour sessions in mere minutes. Complaints surged in late March 2026, with users describing workflows being interrupted and a general sentiment that Claude had become "pretty much useless for serious use" due to these limitations.

The Backdrop: Performance Complaints and Model Evolution

Anthropic's policy move occurs against a backdrop of vocal user dissatisfaction. A widely shared GitHub post from a senior director at AMD in February 2026 claimed, "Claude has regressed to the point it cannot be trusted to perform complex engineering." This sentiment fueled online speculation that Claude was being deliberately "nerfed"—either to control escalating compute costs or to redirect resources toward Anthropic's next-generation "Mythos" project, a security-focused model.

Anthropic denied redirecting compute resources but acknowledged changing its pricing and usage system to better reflect actual customer patterns. A spokesperson told The Information that under the old system, some users hit limits that interrupted work, while others didn't fully use their paid capacity. However, the transition appears to have caused significant friction for power users.

Enter Opus 4.7: A Direct Response to Developer Needs

Seemingly in direct response to these complaints, Anthropic released Claude Opus 4.7 in mid-April 2026. The new model is explicitly tuned for advanced software engineering and "agentic" use cases. A key technical upgrade is the introduction of an "xhigh" (extra high) effort level between the existing "high" and "max" settings, giving developers finer control over the trade-off between reasoning depth and latency on difficult problems.

Anthropic recommends starting with "high" or "xhigh" effort for coding tasks. The model also features "substantially better vision" with higher image resolution and improved performance on professional creative tasks like designing interfaces and slides. Furthermore, Opus 4.7 boasts better file system-based memory, allowing it to retain important notes across long, multi-session projects, thereby reducing the need for repetitive context.

New Controls and a Predictable Release Cadence

Beyond the core model, Anthropic is rolling out new user controls. The company is testing a "task budgets" system, giving developers more precise management over how Claude reasons through lengthy tasks. This directly addresses concerns about inefficient token usage. Accompanying these changes in the Claude Code desktop app is a new /ultrareview command for Max subscribers, which runs a dedicated code review session.

Notably, Anthropic has established a rapid and predictable two-month upgrade cycle for its Opus model line. Opus 4.7 arrived two months after 4.6, which itself came two months after 4.5. This accelerated pace suggests a commitment to continuous, incremental improvement based on user feedback, moving away from the longer gaps seen between earlier versions like 4.1 and 4.5.

Why This Policy and Model Shift Matters

Anthropic's dual moves—clarifying CLI access and launching Opus 4.7—signal a strategic pivot to better serve its developer and power-user base. The CLI confirmation acknowledges that professional use cases extend beyond chat interfaces and into integrated, automated environments. The enhanced controls in Opus 4.7 (effort levels, task budgets) directly tackle the pain points of unpredictable performance and cost overruns reported by users.

For the AI market, this episode highlights the growing pains of commercializing frontier models. Balancing cost, performance, and user experience is a complex act. Anthropic's relatively transparent response, featuring concrete model improvements and policy clarifications, sets a precedent for how AI companies might handle similar backlash. It underscores that for developers, reliable API access and predictable performance are just as critical as raw model capabilities.

The company also noted that Opus 4.7 will be used to test new guardrails against cybersecurity attacks, with lessons applied to the eventual broad release of its Mythos-class models. This indicates Anthropic is using its current flagship model as a testing ground for the safety features required by its most powerful future systems.