vs
Cheap tokens are one axis. Control is the other.
DeepSeek, Qwen, Kimi, and GLM sell genuinely cheap tokens. ProxyLLM is a gateway, not a model, so it does not undercut their price. It puts one endpoint, scoped keys, and hard budgets around whichever providers you use, and runs OpenAI-bound bulk on your ChatGPT subscription at a flat rate.
Free for 7 days, then $129/month. Bring your own model keys. No inference markup.
ProxyLLM vs Chinese model APIs, line by line.
At a glance
ProxyLLM
What the cheap providers do well.
DeepSeek's reasoning and code models are strong and inexpensive. Qwen's open-weight lineup is broad. Kimi and GLM push long context at low prices. For a single workload where the only thing that matters is the per-token bill and the data is fine to send, going direct to one of these APIs is a reasonable call, and ProxyLLM does not pretend to beat their token price. It is a gateway; putting it head to head with a model is comparing the booking layer to a single fare.
The axes that surface later.
Where the data goes
Every external API processes the prompts you send it. For some client contracts, the provider's jurisdiction is a compliance question you have to answer, not a detail.
The price can move
Per-token pricing is a meter someone else sets. A cheap rate today is not a fixed cost you can quote a client for the quarter.
Account and access risk
Sign-up, billing, and availability for some providers carry reliability and geopolitical uncertainty a production system has to plan around.
Provider sprawl
One key and SDK per provider becomes its own tax: separate dashboards, separate logs, no single place that says what was spent where.
It governs them, it does not replace them.
ProxyLLM is an OpenAI-compatible endpoint. Route to DeepSeek, Qwen, and the rest with your own keys behind the same base URL your OpenAI traffic already uses, often as a cheap fallback lane on a routing key. Each client, app, or workflow gets its own key and monthly cap, so a cheap model on a runaway loop still hits a ceiling instead of a surprise invoice. One request log shows what ran, on which provider, for how much. And the work that has to be OpenAI runs through Codex Hosted on your own ChatGPT subscription at a flat cost, while the cheap models carry the lanes they are good at, under a cap. It will not make a DeepSeek token cheaper than DeepSeek sells it. It turns a pile of separate provider keys into one endpoint with per-client budgets and a predictable OpenAI cost.
Run your AI workloads on your ChatGPT subscription.
ProxyLLM runs OpenAI's Codex for you, signed in with your own ChatGPT account. Your apps call one OpenAI-compatible endpoint and the work bills to your flat plan instead of per-token API pricing.
Keep the cheap lanes. Add the control.
Route the workloads that fit a cheap model to a cheap model, under a cap, behind one endpoint with one request log, and put the OpenAI share on a flat subscription.