This page explains the independent MiniMax M3 relay onboarding path on minimaxm3.online. It is designed for buyers who want a public route from evaluation to implementation, while keeping official MiniMax model and Token Plan references close at hand.
Last updated: June 10, 2026
Self-serve API key generation inside the dashboard
OpenAI-compatible chat completions route for quick SDK reuse
Shared credit balance across Playground and API calls
Supports MiniMax M3 and the older MiniMax series through one relay layer
Relay Snapshot
Endpoint
https://minimaxm3.online/api/v1/chat/completions
Key management
/settings/api
Example Request
Launch-readycurl https://minimaxm3.online/api/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer mxm3_your_key" \
-d '{
"model": "minimax/minimax-m3",
"messages": [
{
"role": "user",
"content": "Summarize this product spec and list the implementation risks."
}
]
}'Generate your key
Relay keys are created inside the logged-in dashboard and shown only once. Store them in server-side environment variables, not browser code.
Models you can request
MiniMax M3
Long-context coding, multimodal workflows, benchmark-driven evaluation
MiniMax M2.7
High-capability general text tasks where you still want MiniMax family coverage
MiniMax M2.5
Balanced general workloads and compatibility-oriented routing
MiniMax M2
Legacy or lower-tier MiniMax integration paths that still need one access layer
API evidence snapshot
This section keeps the most citable model facts in one compact table so evaluators do not need to reconstruct them from the homepage.
| Signal | Value | Why it matters | Source |
|---|---|---|---|
| Context window | 1M | MiniMax positions M3 as a 1M-context model for long-code, long-document, and long-session work. | MiniMax model page |
| SWE-Bench Pro | 59.0% | MiniMax reports a 59.0% SWE-Bench Pro score and uses it as a core coding-performance signal. | MiniMax launch report |
| Multimodal training tokens | 100T | MiniMax describes M3 as natively multimodal with 100T multimodal training tokens. | MiniMax launch report |
| Prefilling acceleration | 9x+ | MiniMax reports more than 9x prefilling acceleration using MiniMax Sparse Attention. | MiniMax launch report |
Why buy through this site
If a user already knows they want MiniMax, the main friction is rarely model awareness. It is onboarding friction: where to start, which docs matter, which model ID to use, how to test quickly, and who to ask when the first request fails.
This page now works as real onboarding instead of a contact gate. Users can move from pricing to key generation to first request without waiting for a manual handoff, while still keeping one relay surface for MiniMax model access.
Step 01
Sign in, top up credits, then open the API Console in your dashboard. API usage and Playground usage consume the same credit balance.
Step 02
Generate a self-serve relay key from /settings/api. The key is shown once, so copy it immediately and store it in your backend environment variables.
Step 03
Use the sample payload on this page, test a real call, then move into your own product flow with the same MiniMax model IDs.
API FAQ
The API page explains how to move from evaluation to implementation by buying credits, generating a relay key, and using an OpenAI-compatible endpoint.
No. minimaxm3.online is an independent onboarding and relay layer that points buyers to official MiniMax documentation and gives them a simpler public route into testing and setup.
The public API page positions MiniMax M3 first and also lists MiniMax M2.7, M2.5, and M2 as supported family options through the same relay surface.
Final call
Start by buying credits and generating a key in the API Console. If you still need to validate model behavior first, use the Playground and come back here after one real test.