MiniMax Drops State-of-the-Artwork AI Agent Mannequin—Then Quietly Modifications the License – Decrypt

MiniMax M2.7 is right here, the weights are on Hugging Face, and it is legitimately competing with one of the best closed fashions out proper now.

The numbers: 56.22% on SWE-Professional (a benchmark for software program engineering duties), almost matching Claude Opus 4.6; 57.0% on Terminal Bench 2. An ELO of 1495 on GDPval-AA (a benchmark for real-world information work duties throughout jobs). For context: That is the best amongst open-weight fashions, solely barely beneath Opus 4.6, Sonnet 4.6, and GPT-5.4.

It is a 230B-parameter Combination of Consultants mannequin with solely 10B energetic per inference move, so that you get frontier-level output with out paying frontier-level compute. MiniMax stated it was the primary mannequin to take part in its personal growth—an inner model ran 100+ autonomous rounds of self-optimization, rewrote its personal scaffold, and got here out 30% higher. No human within the loop.

Then the license modified, and the group misplaced it

However shortly after the weights dropped, the Chinese language AI lab MiniMax quietly up to date the phrases: industrial use now requires written authorization from MiniMax.

Non-commercial use stays free and unrestricted. Analysis, private tasks, fine-tuning to your personal setup—none of that modified. However when you’re operating a hosted service or constructing a industrial product, you are in “wants authorization” territory now.

Hacker Information and a Hugging Face dialogue thread stuffed up quick with builders calling it out. The particular friction level is that this: MiniMax is labeling the license “MIT-style,” however MIT permits industrial use by definition. Calling it “Modified-MIT” whereas proscribing industrial use is, to place it charitably, complicated.

Ryan Lee, MiniMax’s Head of Developer Relations, posted an in depth response relatively than the standard company non-answer. His clarification: bad-faith internet hosting suppliers had been deploying degraded variations of earlier MiniMax fashions—unsuitable templates, aggressive quantization, generally not even MiniMax’s precise mannequin—then letting customers stroll away pondering MiniMax ships mediocre work.

“They stroll away pondering MiniMax is mid,” Lee wrote. “We get the reputational invoice, the consumer will get a nasty expertise, and the intense internet hosting suppliers who do the work correctly get drowned out within the noise.”

“A totally permissive license meant we had no strategy to push again on any of that,” he added. “If the license has edge instances that harm reliable group use, inform us. We might relatively repair the textual content than defend it.”

This suits a wider sample. MiniMax constructed its developer popularity on absolutely open releases—M2 below MIT in October 2025, M2.5 below the identical phrases in February 2026. M2.7 is the primary break from that streak, and it got here simply months after the corporate listed on the Hong Kong Inventory Trade in January 2026, elevating round $620M with Alibaba and Abu Dhabi’s sovereign wealth fund amongst its backers.

Different Chinese language corporations, which dominate the open-source area, are additionally testing the waters of close-sourcing AI. Alibaba’s Qwen crew reportedly shifted towards proprietary growth after senior management departures, in accordance with the Monetary Occasions, Xiaomi additionally launched its new MiMo v2 fashions below an in depth supply license. The shorthand that Chinese language labs are open and U.S. labs are closed now not holds.

For these inquisitive about utilizing it commercially, Lee says the authorization course of can be quick and affordable.