MiniMax tests fast coding model with room for a million tokens

Chinese AI company MiniMax has launched M3.1 Flash Preview, a faster and lighter model in the company's M3 series. According to a database page, it was announced for the MiniMax Code tool on September 27, 2026, and made available in the Token Plan subscription the same day. It is a preview, meaning an early version and not a full launch.

The model appeared in the model picker in MiniMax Code before MiniMax itself mentioned it. In the days before, developers had been testing a free model called "Space Bunny" and guessing that it was M3.1.

What the model can do

The model is described as suited to coding, tool use, structured tasks and what is known as agentic reasoning. That means the AI can plan and carry out multiple steps in sequence on its own. It accepts text, images and video as input and responds with text.

The context window is 1 million tokens, according to eesel.ai. That is the amount of text the model can hold in its head at once. For the M2 family, the figure is roughly 205,000.

Reasoning is always on. Users can choose how much the model should think: low, medium, high, xhigh or max, with max as the default. Selecting "none" produces an error.

On the first day, testers measured roughly 90-110 tokens per second, according to eesel. There is no official figure for speed.

Pricing and openness

The model is available only through Token Plan, which costs roughly $20-120 per month, and through MiniMax Code. Pay-as-you-go use is not an option. No per-token price has been published for Flash, and the pay-as-you-go price list covers only M3 and M2.x. Image and video input is listed at $1 per million tokens.

There are no open weights, no model card on Hugging Face and no technical report. Open weights mean that others can freely download and run the model itself.

The model's size has not been disclosed either. According to getdeploying.com, the sister model M3 has 428 billion parameters, of which roughly 23 billion are active per token. eesel cautions against applying the 428 billion figure to Flash without confirmation from MiniMax. According to the same site, M3 costs $0.30 per million input tokens and $1.20 per million output tokens.

All of the information comes from secondary sources, not from MiniMax's own announcement.