MiniMax Launches M3.1 Flash With Speed and Cost Claims
By @sharedot · · 8 pages
- AI
- Minimax
- Model Release
- China AI
Chinese AI startup MiniMax launched updated model M3.1 Flash, claiming 3x faster inference, 70% lower costs, stronger agentic and coding performance.
What Happened
The announcement came in a company statement carried by Reuters, and the news was republished by outlets including U.S. News, The Standard in Hong Kong and several US regional papers.
The Standard notes MiniMax trades in Hong Kong as stock 0100, underscoring that the startup is now a listed company rather than a purely private lab.
The update centers on speed, cost and capability gains across the areas users most often put small fast models to work on: coding, agentic workflows and security tasks.
Why the Claims Stand Out
The efficiency numbers are the eye-catching part. MiniMax says the updated model's online inference speed has nearly tripled while usage costs have been reduced by 70% — claims that, if borne out in practice, would put significant pricing pressure on rivals serving fast, lightweight models.
Capability claims matter just as much: the company says the update brings "generational" improvements in agentic tasks, programming and cybersecurity, the three workloads that dominate real-world adoption of quick-response models.
Because these are the company's own claims rather than independent benchmark results, buyers will want third-party evaluations before drawing conclusions.
How It Fits the Week
The release lands in a crowded stretch for open and fast models. MiniMax's push on inference speed and cost suggests Chinese labs are competing as much on economics as on raw capability — a differentiator that matters for developers running high-volume agentic workloads.
The Standard also reports MiniMax's revenue nearly quadrupled in the first half as AI demand surged, giving the company momentum behind the update.
The Evidence So Far
What is confirmed now: the launch itself, the company's stated improvement areas (agentic tasks, programming and cybersecurity), and its performance and cost figures — nearly tripled online inference speed and a 70% reduction in usage costs.
Reuters reporting attributes the figures directly to MiniMax's statement, and all four pack outlets carry the same core language.
What is not yet available in the pack: independent benchmarks, head-to-head comparisons with DeepSeek, Meta's Muse or Google's Gemini models, or details on how the improvements were achieved. Readers should treat the numbers as vendor claims until evaluations appear.
The Stakes
A 70% usage cost cut would compound quickly at scale, and The Standard's report of quadrupled first-half revenue shows MiniMax has commercial traction to protect.
The update also signals intensifying US–China competition in efficient models, arriving the same week the feed covered Washington cutting federal ties with Anthropic and hyperscaler AI debt concerns from S&P Global. Efficiency, not just frontier capability, is emerging as a competitive battleground.
What Comes Next
With Meta releasing Muse Spark and Muse Wonder models, Google shipping Gemini 4 Argon, and OpenAI launching Astra in recent weeks, the fast-model tier is now one of the most contested segments in AI. M3.1 Flash's update keeps Chinese labs squarely in that race.
Sources
- centraloregondaily.com › China's MiniMax launches updated AI model M3.1 Flash, claims major improvements
- money.usnews.com › China's MiniMax Launches Updated AI Model M3.1 Flash, Claims Major Improvements
- thestandard.com.hk › China's MiniMax launches updated AI model M3.1 Flash, claims major improvements
- lufkindailynews.com › China's MiniMax launches updated AI model M3.1 Flash, claims major improvements