Fireside Chat, Interview
MiniMax: Scaling Intelligence for Everyone | Linda Sheng | RAISE Summit 2026
- Minimax has evolved from a small AI startup to a global public business, recognized as a top video model and text-to-speech company.
- The company released its M3 large language model in June, which is widely used globally as an open-source solution.
- M3 is the first open-source model natively trained on text, video, images, and unstructured multi-modal data from pre-training.
- The M3 model supports a 1 million token context window, matching the state-of-the-art capabilities of OpenAI and Anthropic models.
- Enterprises are currently categorized into three generations of AI readiness based on daily token consumption: millions, billions, or nearly a trillion.
- Token consumption is projected to grow by 10 to 100 times in enterprises adopting "co-work" use cases like document summarization and PowerPoint generation.
- CIOs and enterprises shifted toward diversified model strategies in Q2 due to budget constraints, specifically citing OpenClaw's impact on accelerating the race to the agent economy.
- Many Fortune 500 companies, particularly in security-sensitive banking sectors, are increasingly demanding open-source models to avoid vendor lock-in.
- European AI startups are highly active but dispersed, with notable success in Sweden (e.g., Lovables, Lagoras) and Western Europe (e.g., Freepik/Magnific).
- Minimax advises AI founders to build internal evaluation benchmarks to test new state-of-the-art models within 24 hours of release.
- The company operates a feedback loop where startups can influence model development, with new model iterations released approximately every 1.5 months.
- Minimax's internal operations utilize 20 AI agents to handle work traditionally assigned to interns and junior staff, making agents integral to the workflow.
- The company predicts narrow-sense AGI—where AI outperforms humans in knowledge-intensive work—is approximately three years away.
- Finance and legal sectors are identified as the next major areas for AI disruption following coding, due to their reliance on structured text and rules.
- Healthcare is expected to remain a later disruption target due to the highly specialized nature of its knowledge requirements.
- Emerging "AI-native" employees are becoming multi-disciplinary, capable of performing roles across law, finance, and coding without traditional specialized training.
- Minimax's "10x Team" currently employs human experts to review AI work in finance to address hallucinations in specialized domains.