MiniMax is one of China’s leading AI labs. Its models can create video, music, text and code and act as agents to execute complex, multi-step tasks. MiniMax charges customers to access its models, but in many cases also lets them run and modify the underlying ‘weights’ on their own equipment without charge. “It’s a catalyst for… innovation,” the firm’s co-founder and chief executive, Dr Yan Junjie (IO), tells Scottish Mortgage manager Lawrence Burns. In this episode of the Trust’s Invest in Progress podcast, the two discuss how MiniMax has turned constraints to its advantage and ultimately hopes to change our understanding of the universe.
Background:
Scottish Mortgage invested in MiniMax when it floated in Hong Kong in January 2026.
If you’ve ever watched AI-generated videos on YouTube, there’s a good chance they were generated using its H-series AI models. The firm’s M-series models focus on text-based tasks, including coding, reasoning and complex problems that benefit from an ‘agentic’ approach.
MiniMax charges customers to access these models, but in many cases also lets them run and modify the underlying ‘weights’ on their own equipment without charge. “It’s a catalyst for… innovation,” IO explains.
In this interview, the founder-CEO also explains why the firm’s efforts to minimise its inference costs – the cost of running its models – could give it a long-term advantage in achieving ‘artificial general intelligence’ (AGI). And why that could be the key to accelerating progress in scientific discovery. In addition, he shares his view on the differences between the broader Chinese and US AI ecosystems, and why he believes their paths won’t converge.
Timecodes:
00:03 Coming up…
01:14 Introduction
03:34 Dr Yan Junjie (IO) interview begins
04:10 Minimising inference, maximising intelligence
05:15 Defining AGI
05:50 MiniMax’s M- and H-series AI models
08:11 Self-learning in Henan province
10:23 Mentored by Horizon Robotics’ Dr Kai Yu
13:50 Glow and the ChatGPT moment
16:44 MiniMax’s multimodal AI strategy
18:21 Early investors
19:45 Mavis and agentic AI
21:27 Low inference costs as a ‘moat’
23:30 Closing the gap with US models
26:09 Focusing on efficiency
27:42 ‘The largest post-training system’
29:12 Open source as a catalyst for monetisation
33:20 Culture shaped by constraints
36:02 Chip-agnostic architecture
37:28 AI ecosystems: China v US
39:09 Competition within China
41:18 A deeper understanding of the universe
44:14 Lawrence Burns on the investment case
54:43 Podcast lookahead
Check the podcast description to ensure this content is suitable for you. Your capital is at risk.
Presenter: Claire Shaw
Executive Producer: Leo Kelion
Line producer: Jessica Rooney
Broadcast Technician: Samuel O’Hare
Editor: Jody Black
","author_name":"Scottish Mortgage"}