The Bottom Line: Users can now use GPT-5, Gemini, open-source models via OpenRouter, local models, or enterprise gateways as backends for Claude Cowork and Code in Claude Desktop.
Anthropic has enabled Claude Cowork and Code in the desktop application for use with arbitrary language models. The feature was rolled out without formal announcement, blog post, or press release and is available only in the technical documentation.
Anthropic has unlocked support for third-party inference in Claude Cowork and Code. Via Claude Desktop, users can now leverage any language model as the backbone: commercial models like GPT-5 or Gemini, open-source variants via OpenRouter (such as GPT OSS 120B), locally running models on their own machine, or enterprise solutions like AWS Bedrock, Google Vertex, and other foundries.
The rollout occurred without public announcement, without a blog post, and without official press release. Instead, Anthropic documented the feature only in the technical documentation. A security researcher discovered the feature by chance; over 20 hours later, there was still no official statement from Anthropic.
For practitioners, this means: those who want to use Claude Cowork and Code without being locked to Anthropic’s own models can now leverage the tools with local or OpenRouter-sourced models — for example, through the LiteLLM proxy for compatibility. This opens flexibility in model selection without sacrificing Cowork and Code functionality.
Source: ainews-dev.lumi-systems.io · Published May 3, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.5.2.