MiniMax is a multimodal model platform covering text, voice, and video. Many teams use it for customer service, content generation, and voice interaction.

This guide explains how to obtain access, call the API, and integrate it without assuming that every capability uses the same protocol.

Note: MiniMax is a trademark of its owner. This article reflects common versions as of October 2026; interfaces and prices are subject to official or in-site documentation.

1. Common MiniMax API Capabilities

  • Text chat, writing, and summarization.
  • Speech synthesis and speech recognition, depending on the model.
  • Video generation, depending on the model.
  • Agent and tool calling.
  • Multimodal input.

Capabilities vary widely between models, so check the current model list and parameters before designing a workflow. A model that handles text well may not support vision, and a voice model may have entirely different request and response shapes.

Build a capability matrix for the models you plan to use. List input types, output types, maximum file size, and whether the call is synchronous. That matrix prevents the common mistake of designing one client abstraction for interfaces that do not share semantics.

2. Getting a Key

Official MiniMax

Register on the MiniMax open platform, create a key, and top up. This route suits teams that can use the official platform directly.

ZSQH AI Gateway

  1. Register for the ZSQH AI Gateway.
  2. Create an API key and set a quota.
  3. Top up the account.
  4. Choose the model and interface according to the in-site documentation.

Note that not every MiniMax capability uses an OpenAI-compatible protocol. Text models usually do. Voice and video may use task-based interfaces that require separate adaptation, so confirm the endpoint before you reuse an existing chat client.

3. Text Model Call

If the gateway provides an OpenAI-compatible endpoint:

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-key",
    base_url="https://ai.zhongshangqihang.com/v1",
)

resp = client.chat.completions.create(
    model="minimax-text",
    messages=[{"role": "user", "content": "Reply with OK only"}],
)
print(resp.choices[0].message.content)

Model names follow the model pricing page. Use the smallest request possible for the first test; it separates key and endpoint problems from prompt or parameter problems.

Enable streaming if the product needs partial results. The request shape stays the same, but the client must handle incremental events and finalize the message when the stream ends. Set a timeout that matches the model's expected latency.

4. Voice and Video Interfaces

Voice and video are usually asynchronous tasks:

  1. Submit a task and receive a task ID.
  2. Poll the status.
  3. Download the result when it succeeds.

These interfaces differ from OpenAI Chat Completions and cannot be used through a chat SDK without adaptation. Before integration, confirm:

  • Whether asynchronous tasks are supported.
  • The callback or polling method.
  • File size and duration limits.
  • How long results are retained.

Build a job record for each submission so that a timeout does not cause a duplicate task, and download finished files before their URLs expire.

5. Integration Test Checklist

  1. Text request and streaming.
  2. Tool calling, if the model supports it.
  3. Voice or video task submission and polling.
  4. Long-text and large-file limits.
  5. Concurrency and rate limits.
  6. Bill reconciliation.

Run the checklist with production-like inputs. A short text prompt and a long audio file exercise very different parts of the platform, and a test that only covers the happy path will miss the limits that matter at launch.

6. Common Issues

Symptom Cause Action
401 Wrong key or endpoint mismatch Check the key and Base URL
404 Incorrect interface path Follow the official or in-site documentation
Model unavailable Model name or permission Check the pricing page and enablement status
Task stays processing Queueing or file too large Reduce the size and retry later
429 Rate limit Lower concurrency and retry

7. Cost Control

  • Text: billed by input and output tokens.
  • Voice: billed by duration or character count.
  • Video: billed by second, resolution, or task.
  • Failed retries also generate cost.

Start with a small test, confirm the unit price and success rate, and only then run a batch. For voice and video, track cost per accepted output rather than cost per request, because failed tasks and retries can change the economics significantly.

FAQ

Is the MiniMax API OpenAI compatible?

Text usually has a compatible interface. Voice and video are mostly independent task interfaces, so follow the official documentation.

Is access stable from mainland China?

MiniMax is a domestic platform and is generally more stable than overseas models. Your specific network and region still matter.

Can I connect Claude and GPT at the same time?

Yes. With a unified gateway, one key can call multiple models, which simplifies comparison and fallback.

Next Steps

If you need a unified entry point, pay-as-you-go billing, and multi-model switching, you can register for the ZSQH AI Gateway and run small tests against the text, voice, and video capabilities. Model names and prices are listed on the model pricing page.

Related reading: