newsfilter.io
Fireside Chat, Interview

MiniMax: Scaling Intelligence for Everyone | Linda Sheng | RAISE Summit 2026

  • Minimax plans to release a new model every 1.5 months, utilizing user feedback to refine subsequent generations for specific use cases while expanding the M3 open-source model with native training on text, video, image, and unstructured data.
  • M3 is projected to leverage a 1-million context window and native vision-language model (VLM) capabilities to enable simultaneous reasoning and visual processing, unlocking significant opportunities across multiple media sources.
  • Enterprise token consumption is expected to increase by 10 to 100 times as "co-work" use cases expand beyond coding to include PowerPoint summaries, document summarization, and general office work.
  • Enterprises are projected to adopt diversified model strategies and internal routing platforms with credit systems to mitigate costs associated with closed-source providers and ensure data security.
  • Minimax aims to increase its European user base by supporting regional founders, anticipating that entrepreneurial success in the region will be driven by creative AIGC companies and dispersed individual founders building global startups.
  • Users are advised to benchmark new models against specific workloads within 24 hours to determine suitability, while the workforce is expected to evolve into "AI native" or "AI enhanced" multi-disciplinary roles that may outperform specialized experts.
  • Fresh graduates who leverage AI to cross-check information are anticipated to become top-performing employees, and narrow-sense AGI where AI performs knowledge-intensive work better than humans is projected to be three years away.
  • Finance is identified as a top target for AI disruption due to its reliance on rules and numbers, while legal is expected to be the next significant use case; however, healthcare disruption is anticipated to occur much later due to high specialization.
  • AI integration in specialized knowledge areas will likely involve human expert review teams to mitigate hallucinations, potentially transforming AI from a worker into an expert and enabling full automation of tasks via agents.
  • Minimax anticipates that AI agents will become an integral part of its team, while the industry will continue a "winnowing out" process where new state-of-the-art models emerge quarterly, potentially forcing traditional workflows to be replaced by agents.