Show HN: Claude-Bridge – Use GPT, Gemini, and Other LLMs with Claude Code

3 points

a year ago

A few days ago, I got curious about how Claude Code works internally. I set up a man-in-the-middle proxy and started tracing requests. That culminated in a little tool to visualize those requests called claude-trace:

https://nitter.net/badlogicgames/status/1929312803799576757#...

Now that I had full control over requests made by Claude Code, the next obvious step was to make Claude Code talk to other LLM endpoints, which led to claude-bridge. Here are a few fun little test runs with GPT 4.1, Gemini 2.5 Flash, Grok 3, and LLama 4.

https://nitter.net/badlogicgames/status/1930090999004443049#...

There are, of course, limitations, as outlined in the README.md in the repo linked to in this submission. But it's a fun little tool to quickly test agentic coding assistant capabilities and tool use in real-world scenarios using Claude Code's fantastic interface. Gemini fared pretty well; the rest, not so much.

Nothing came close to Sonnet and Opus, which I attribute in part to the prompts sent by Claude Code being tailored towards the capabilities of Anthropic's models. Claude Code's pricing is also really hard to beat for this token-heavy use case.

1 comment