Every MiniMax model on one balance: text, voice and video
YiduoChan is a hosted gateway for MiniMax's whole catalogue. One API key and one prepaid balance reach the M-series language models, Speech 2.8 with voice cloning, and H3 video generation. Point any OpenAI-compatible client at https://yiduochan.com/v1; Anthropic-compatible clients such as Claude Code use https://yiduochan.com/v1/messages and work the same way.
Why one balance matters
Most gateways carry language models only. A product that narrates its own articles, or turns a script into video, ends up holding one account for text and a separate one for the rest — two sets of credentials, two invoices, two balances to keep topped up, and no single place to see what the whole feature cost. Here a chat completion, a cloned-voice narration and a rendered clip all draw on the same balance and land in the same itemised usage log.
Why MiniMax
MiniMax is a Chinese AI lab whose M-series language models are built for agentic and coding work rather than chat: long tool-calling loops, repeated file reads, and the kind of session where the same context is re-sent hundreds of times. That shape is what makes agentic coding expensive, and it is the shape these models are priced and tuned for. Alongside them the same lab ships current-generation speech and video models, which is why one vendor can cover all three modalities at all.
This site is a gateway, not the model vendor. The models, their quality and their rate limits are MiniMax's; what the gateway adds is the single balance across every modality, two API surfaces instead of one, and a way to pay that does not assume a human filling in a signup form.
Two compatible surfaces
The OpenAI-compatible surface at https://yiduochan.com/v1 serves chat completions, speech and video, so existing OpenAI SDK code works by changing the base URL and the model name. The Anthropic-compatible surface at https://yiduochan.com/v1/messages is what Claude Code and other Anthropic-API tools talk to. Both draw on the same balance; neither requires a separate contract.
Models
- MiniMax-M3 — text: long-context reasoning, a one-million-token window
- MiniMax-M2.7 — text: general chat, agent loops and coding
- MiniMax Speech 2.8 — voice: speech in many languages, and cloning a voice from a sample
- MiniMax H3 — video: text to video, image to video
Current rates and context limits are on the pricing page and the MiniMax model overview.
Integration guides
Getting a key
People create an account and top up by card. Autonomous agents do not need an account at all: pay USDC over the x402 protocol at /api/x402/credits/5 and the response returns a working API key. There is also a pay-per-call endpoint at /api/x402/chat/completions, where an unpaid request answers 402 Payment Required with the terms in a header and a signed request returns the completion itself. Wallet sign-in and direct USDC deposits are documented in the agent guide (markdown), with machine-readable details in the agent manifest and llms.txt.
How billing works
Credits are prepaid and spent per token or per character as requests are served; there is no subscription and no monthly minimum. A request that fails upstream is not charged. Every call is itemised in the console usage log with the model, the tokens counted and the amount deducted, so a bill can be reconciled against the requests that produced it. Published rates and the exact billing units for each model are on the pricing page.