Developer builds multi-model gateway compatible with OpenAI and Anthropic APIs

On October 24, 2026, a developer posted on V2EX about a multi-model gateway they built themselves. The motivation was straightforward: they mainly code with OpenCode / Claude Code, and the monthly subscription and API costs were adding up, while each model required a separate platform and a pile of keys to switch between. Over several months of evenings, they built a small project to solve this.

The gateway accepts a single key and routes requests to DeepSeek, GLM, Kimi, Claude, GPT, and other models. It is compatible with both the OpenAI and Anthropic API formats, so in OpenCode, Claude Code, and Cursor, users only need to change the base URL to start using it. When friends said they did not want to configure a client, the developer added a web chat interface modeled on the Claude web experience, where after signing in you can chat with each model directly and switch between them to compare outputs.

The post highlights several lessons from building it. First, the streaming return formats of the OpenAI and Anthropic APIs differ significantly, and Claude Code is quite picky about format, so it took a long time to stabilize. Second, usage is billed per token in real time; handling duplicate billing caused by interrupted streams and retries turned out to be more work than expected. Third, for login, the developer kept only Google and Telegram sign-in, skipping email verification altogether. Fourth, top-ups currently accept only cryptocurrency, with a minimum of $1 and no expiration on balances; domestic payment methods are not yet supported.

That is where the project stands at present, based on the available material.