If you’re using a budget Anthropic plan, you’ll likely hit your daily or weekly quota while working on Claude coding projects. To keep the momentum going, you can switch to a local open-source model. The article suggests checking your usage with the command `/usage` and recommends GLM-4.7-Flash from Z.AI or Qwen3-Coder-Next as solid options. For those short on disk space or GPU memory, smaller quantized versions are available, though they may sacrifice some quality. The author promises a follow-up post on selecting the best open-source model based on your specific needs. For connecting to a local model, the article outlines two methods. The first is through LM Studio, which is user-friendly for running open-source LLMs and vision models. With version 0.4.1, you can connect to Claude Code. The steps include installing LM Studio, starting a server on port 1234, and configuring environment variables to link Claude Code to LM Studio. Users should manage their expectations regarding speed and performance when using this setup. The command `/model` can help confirm which model is currently in use. The second method involves connecting directly to llama.cpp, the open-source project that LM Studio is built on. Although this approach is possible, the author suggests that LM Studio is generally faster and easier unless you have specific needs or are fine-tuning a model. While the local option serves as a backup, expect reduced speed and code quality, especially on less powerful machines. Switching back to Claude when your quota resets is straightforward, allowing you to continue coding without interruptions.