MiniMax M2.5 promises "intelligence too cheap to meter" as Chinese labs squeeze Western AI pricing
Key Points
- Chinese AI company MiniMax has released M2.5, an open-weights model under MIT license that can autonomously plan and execute tasks without constant human guidance.
- In benchmarks, M2.5 scores 80.2% on SWE-Bench Verified for programming and outperforms Claude Opus 4.6, GPT-5.2, and Gemini 3 Pro on web search and office tasks, at ten to twenty times lower cost.
- The lightweight M2.5-Lightning variant generates 100 tokens per second, with MiniMax claiming that one hour of continuous operation costs just one dollar.
Chinese AI company MiniMax out of Shanghai has released its new open-weights model M2.5 under the MIT license.
MiniMax says M2.5 was trained using reinforcement learning across hundreds of thousands of complex environments, which should help it deliver strong results with low token consumption, especially on long-running tasks.
The model learned to optimize its actions through its own planning, setting it apart from models that only plan when prompted to do so. It can also handle Word, Excel, and PowerPoint files quickly and smoothly, according to the company.
Based on the published benchmarks, the model posts top scores in programming: 80.2 percent on SWE-Bench Verified, 79.7 percent on the Droid platform, and 76.1 percent on OpenCode, in some cases edging out Anthropic's Claude Opus 4.6. MiniMax says M2.5 also beats GPT-5.2, Gemini 3 Pro, and Claude in web search (BrowseComp: 76.3 percent), AI agent tool use (BFCL: 76.8 percent), and office tasks like Excel work.
Cheap and fast
MiniMax ships two variants: M2.5-Lightning pushes 100 tokens per second, making it twice as fast as other top models. One hour of continuous operation costs just one dollar, according to MiniMax. That makes the model ten to twenty times cheaper than comparable offerings from Anthropic, Google, and OpenAI. Four M2.5 instances could run continuously for a full year at $10,000, the company claims.
M2.5 is also freely available on Hugging Face and GitHub. A detailed technical report is up at minimax.io. API access and a dedicated coding plan are available at platform.minimax.io.
"M2.5 is the first frontier model where users do not need to worry about cost, delivering on the promise of intelligence too cheap to meter," the company said. Pricing like this could put more pressure on Western AI labs, which are growing fast but still losing money.
Reportedly, even some US startups are turning to Chinese models because of lower prices, though US enterprise AI remains firmly in the hands of Microsoft, Google, OpenAI, and Anthropic. And even Chinese labs acknowledge that US labs are still ahead, with chip scarcity not helping matters. The picture, in other words, isn't black and white.
Beyond language models, MiniMax builds video AI tools. In June 2025, the company released Hailuo 02, the second generation of its video AI model. Since September 2025, Disney, Universal, and Warner Bros. have been suing the company for copyright infringement tied to its models.
That didn't stop MiniMax from going public in early January 2026. The stock surged 109 percent to HK$345 on its first day of trading. The M2.5 release gave the stock another boost, with shares closing 15.7 percent higher at HK$680. MiniMax was founded in 2021 by former SenseTime employees.
AI News Without the Hype – Curated by Humans
As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.
Subscribe now