220M tokens for $9. Your OpenAI SDK, Claude-class models.
Kimi-K2.6 scores 80.2 on SWE-Bench Verified — Claude Opus 4.6 scores 80.8 — at roughly a tenth of the price. DeepSeek-V4-Flash costs a fifth of that again, and the headline figure is counted in it. One OpenAI-compatible endpoint, plan or pay-as-you-go. Free key in 30 seconds.
400k+ requests served · open source · pay by card or crypto · no KYC
https://router.mingles.ai/v1 Try the API right now
Ask our AI assistant anything — it runs on Kimi-K2.6 through this very API. No signup needed.
What you can build today
Real outcomes, not boilerplate. Pick a flow, copy a config, ship.
New to this? A coding agent is a program that reads your files, runs commands and writes code for you — Claude Code and OpenAI Codex are the well-known ones. They need a model behind them, and that is what we are. Set one up in 3 steps →
A coding agent of your own
The Claude Code / Codex idea, open source: DeepSeek Harness reads your repo, runs commands and writes code — on whichever model you pick.
See config →Vibecoding in Cursor
Code with AI without paying $20/mo. Drop your key into Cursor, pick a model, write apps.
See config →No-code automations in n8n
Build agents that read your email, summarize Notion docs, post to Telegram — all visual, no code.
See config →Always-on bots with OpenClaw
Spin up a Telegram or Discord bot that thinks. No server. No DevOps. One config file.
See config →AI in your IDE with Cline
Free Claude-Code-style agent that opens files, runs commands, edits projects — pointed at one endpoint.
See config →Built for two workflows
For developers & vibecoders
One OpenAI-compatible endpoint in Cursor, Cline, n8n, OpenClaw, Hermes and custom agents. Free key, pay by card or crypto.
Start buildingFor agencies
Control LLM cost across client bots and workflows. Per-client keys, limits, usage tracking and fallback routing.
See agency backendPlug-and-play
Anything that says “OpenAI-compatible” works. One URL, one key, that's the integration.
Cheap, no lock-in
Per-token pricing well below the public-API median. Pay by card or crypto — top up only what you use.
Agents-first
Tool-calling, streaming, request tracing — everything Cursor/Cline/OpenClaw need just works.
What people run on it
I use Mingles Router in my service Sonar to score and analyse posts and comments in Telegram chats. The main advantage for me is the price: at high processing volumes it comes out significantly cheaper than going to OpenAI directly or through OpenRouter, and for tasks like these the model quality is perfectly adequate.
The service can occasionally glitch or respond a little slower, but in our scenario that's not critical — the cost of processing a large volume of messages matters far more to us than minimal latency. For mass LLM analysis it's a very good price-to-quality option.
Started using the Router for vibecoding and so far it's a real win. I dropped it into my everyday tasks — sketching a feature, fixing code, digging into an error, quickly rewriting something.
The best part is not having to think about the cost of every request at all. Even on the free tier the limits last surprisingly long — you can put together a decent chunk of a project in an evening. For this way of working, where you're constantly asking the model something and generating code, it's a very good deal.
Real users, quotes translated from Russian.
One minute, if you do code
The official OpenAI SDK works as-is. Change one variable.
from openai import OpenAI
client = OpenAI(
base_url="https://router.mingles.ai/v1",
api_key="sk-your-free-key",
)
resp = client.chat.completions.create(
model="moonshotai/Kimi-K2.6",
messages=[{"role": "user", "content": "Make me a startup name."}],
)
print(resp.choices[0].message.content)Need Node / curl / Cursor / Cline / n8n? See all configs →
100 % open source.
Read the code, fork it, self-host it. We publish the router under a permissive license.
Frequently asked questions
▸Do I need to know how to code?
No. Most of our users run no-code agents (n8n, OpenClaw) or vibecoding tools (Cursor, Cline). The API key is a single line in a settings screen — that's the deepest you need to go.
▸What apps can I plug it into?
Anything that supports an "OpenAI-compatible API" or "custom base URL": Cursor, Cline, Continue, Aider, OpenClaw, n8n, LibreChat, Open WebUI, Jan, LM Studio, and more.
▸How is this different from ChatGPT?
ChatGPT is a chat app for one user. We are an API — the engine that powers apps and agents. If you only want to chat, use ChatGPT or our /app console. If you want apps and bots to think, you want this.
▸How do I pay?
Pay by card or crypto in the console — no KYC. Pick a monthly plan, pay-as-you-go from your balance, or start free.
▸Where can I see the code?
The router is open source: github.com/MinglesAI/gonka-proxy. Fork it, self-host it, or just read how it works.
Get your free key and start building.
30-second sign-up. Pick your tool. Paste the key. Ship.