MiniMax Ships Open Weights Friday, Oct. 23. Nobody Is Surprised.

Four days. That's how long MiniMax is making you wait between a hosted launch and open weights. M3.1 Flash went live on the API on Monday, Oct. 19 at $0.15 per million input tokens and $0.60 per million output tokens, and the weights land on Hugging Face on Friday, Oct. 23, no waitlist, no sales call, no enterprise tier gating the download.

We covered this exact pattern on Sunday, Oct. 18, when DeepSeek put its fourth V4-line model of the quarter on Hugging Face. MiniMax answered in two days.

This isn't a second data point. It's a trend line pointing at coordination, not coincidence.

What MiniMax actually shipped

MiniMax Group Inc. released M3.1 Flash, a faster and cheaper update to the M3 model it launched in June. Same 428-billion-parameter mixture-of-experts shell, roughly 23 billion active per token, one-million-token context window, and now with video. MiniMax claims a "generational leap" in agentic tasks, coding, computer use and cybersecurity.

Translation: we shipped a cheaper, faster version of last quarter's model and are calling the discount a breakthrough.

The numbers are mixed. MiniMax scores 73.7 on DeepSWE v1.1 against DeepSeek's 77.0. On Automation-Bench v1.0.6, MiniMax scores 62.2 ahead of DeepSeek's 60.5. Artificial Analysis is more measured: M3.1 Flash scored 42 on its Intelligence Index, up 13 points from M3, ties GLM-5.3-Flash at 42, edges Gemini 3.8 Flash at 41, trails DeepSeek V4.1 Flash 1017 at 45. The same firm flagged a 93 percent hallucination rate; the model tries to answer every test question instead of declining when uncertain.

The model ran anonymously on OpenRouter since late September under the name Space Bunny, processing 56.7 trillion tokens, the most of any stealth model there. MiniMax used the public internet as free QA before slapping a name and price tag on it.

The coordination gap

One lab shipping fast is engineering talent. Two labs shipping on overlapping four-day windows, both hosted access first and open weights four days later, both pricing a Flash tier to undercut each other's peak rate, is a story about posture.

Chinese labs are treating release timing itself as a competitive weapon, not just model quality. That's an inference, not a press release. Nobody at MiniMax or DeepSeek has announced a coordinated calendar. But both labs chose the same gap—hosted access first, weights four days later—as if that's now house style across Chinese labs rather than one company's quirk.

If that's right, Western labs aren't losing a speed race. They're losing a race they didn't know had a second runner.

What to watch: whether the Oct. 23 weights ship with a real license or another ambiguous "open-weight" label with no terms. Whether independent benchmarkers replicate the DeepSWE and Automation-Bench numbers once downloadable. Whether any American lab announces a release cadence commitment before year-end.