Intelligence in the Open — Edition 01: Reasoning, Evaluation & Long-Horizon Agents

Intelligence in the Open is a recurring MiniMax research series bringing together AI researchers, engineers, technical founders, and investors for candid conversations about the work shaping the next generation of intelligent systems. For our first edition, we’ll convene a curated group of attendees at Upscale for an evening of short research presentations, open discussion, food, and networking. The initial program will explore structured multi-agent reasoning, process-level error localization, counterfactual evaluation, and new methods for understanding how AI systems behave across extended, real-world tasks. Each researcher will present recent work in a focused 10–15 minute talk, followed by a broader discussion and time to connect directly with the speakers and other attendees working across foundation models, agentic systems, post-training data, multimodality, benchmarks, and evaluations. Speakers Additional speakers, research topics, and event partners will be announced as the program develops. Schedule 6:00–6:30 PM — Arrival and Check-in Meet fellow researchers, founders, engineers, and investors before the program begins. 6:30–6:40 PM — Welcome and Opening Remarks An introduction to, MiniMax, Intelligence in the Open and the evening’s research themes. 6:40–7:40 PM — Research Presentations Three to four focused talks from researchers presenting recent work across reasoning, agents, and evaluation. 7:40–8:00 PM — Open Discussion and Panel Q&A All speakers will return for a moderated group discussion connecting the themes across their presentations. 8:00–9:00 PM — Food and Networking Continue the discussion directly with speakers and connect with others working across the AI research and startup ecosystem. Presented by MiniMax, Protégé and Together AI. Hosted at Upscale. MiniMax is a frontier AI lab building multimodal foundation models and AI products for developers, enterprises, and creators. Its latest model series brings together long-context reasoning, coding and agentic capabilities, and native multimodal understanding for real-world workflows.