MiniMax M3.1 Flash API: Official Endpoint, Model ID, and Pricing

Two different doors

MiniMax documents its own API. Spacebunny does not resell that API and does not put a MiniMax key in the browser. The playground calls OpenRouter with minimax/minimax-m3.1-flash:free from the server. Use the official API if you want the vendor contract; use the playground if you want a metered chat with our credit cap.

Official endpoint and model id

Checked against MiniMax’s text-generation guide and the OpenAI-compatible chat reference. Base URL https://api.minimax.io/v1. Path /chat/completions. Model id MiniMax-M3.1-Flash. The guide also lists an Anthropic-compatible base URL at https://api.minimax.io/anthropic. Context window in that guide: 1,000,000 tokens. An older text/chatcompletion_v2 path is marked deprecated on their docs; prefer the OpenAI-compatible path.

curl https://api.minimax.io/v1/chat/completions \
  -H "Authorization: Bearer $MINIMAX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"MiniMax-M3.1-Flash","messages":[{"role":"user","content":"Say hi in one sentence."}]}'

Pricing they published

The 19 October 2026 research post lists a flat rate per 1M tokens: $0.15 input, $0.60 output, $0.015 cached input. That is MiniMax’s list price, not a Spacebunny invoice. Our $10 pack buys Spacebunny credits. It is not a balance on platform.minimax.io.

OpenRouter, including the date the free route ends

On 21 October 2026 the public models API listed minimax/minimax-m3.1-flash:free with context 1,000,000 and prompt/completion price 0. It did not list minimax/minimax-m3.1-flash. OpenRouter’s model page says the free variant is retired after 22 October 2026 at 12:00 UTC, and that the base slug is then billed at the list price above. Authenticate to OpenRouter with a Bearer key from their dashboard. Spacebunny’s key is not shared.

Errors and limits we will actually state

On the official API, a missing or bad key fails authentication. This page does not invent a catalog of status codes. On the Spacebunny playground, HTTP 429 from upstream is shown as rate limited, 402 as a provider billing problem, and 5xx as a temporary outage. Our own request hard-cap is about 100k tokens, with a completion cap of 2048. Neither cap is MiniMax’s 1,000,000-token window.