Hugging Face Publishes Open-Source Guide After Anthropic Limits Claude

SAN FRANCISCO — Hugging Face published a migration guide Thursday urging developers to switch to open-source AI models after Anthropic began restricting access to its Claude models on open agent platforms for Pro and Max subscribers.

The guide, titled “Liberate your OpenClaw,” targets users of OpenClaw and similar open code-agent tools that had relied on Claude as their default model. Anthropic’s policy change effectively cuts off paying subscribers from using Claude through third-party open agent platforms, forcing developers to seek alternatives or use Anthropic’s own tooling.

“You do not need a closed hosted model to get OpenClaw back on its feet,” the Hugging Face team wrote in the blog post, authored by CEO Clem Delangue and several staff engineers.

Two Migration Paths

Hugging Face laid out two options for affected developers. The first routes users through Hugging Face’s own Inference Providers, recommending GLM-5 from Zhipu AI as the primary replacement model, citing its “excellent Terminal Bench scores.” The company noted that Hugging Face Pro subscribers receive $2 in free monthly credits toward inference usage.

The second option targets developers who want to run models locally, recommending Qwen3.5-35B via llama.cpp — a setup the company said works on machines with 32 GB of RAM and requires no API costs.

“The hosted route is the fastest way back to a capable agent. The local route is the right fit if you want privacy, zero API costs, and full control,” according to the post.

Broader Implications

The episode highlights growing tension between commercial AI providers and the open-source agent ecosystem. Anthropic’s decision to restrict Claude access on open platforms — while continuing to offer it through its own products and API — raises questions about vendor lock-in as agentic AI tools become central to developer workflows.

For developers who built toolchains around Claude’s capabilities, the restriction forces a choice: migrate to open-source alternatives or consolidate within Anthropic’s ecosystem. Hugging Face, which operates the largest open-source model hub, has a clear commercial interest in steering those developers toward its own infrastructure.

The move also underscores the rapid maturation of open-weight models. That Hugging Face can credibly position GLM-5 and Qwen3.5 as functional replacements for Claude in agent workflows suggests the performance gap between proprietary and open models continues to narrow, at least for code-agent use cases.

Neither Anthropic nor Hugging Face immediately responded to requests for additional comment on the policy change and its scope.