MiniMax × OpenAI Dev Night & Hackathon Weekend

Last weekend in San Francisco reminded us why we build.
Across two back-to-back events, nearly 300 engineers, founders, and C-suite leaders came together for a weekend focused on one thing: shipping real systems . No hype cycles, no abstract demos — just builders working through the realities of production AI.

We kicked things off with the MiniMax M2.1 Workshop & Dinner , co-hosted with AI Valley and supported by our friends at Hanwha AI Center . Nearly 100 senior builders packed the room for a high-signal evening centered on deployment, evaluation rigor, and agentic workflows that actually ship .

This was an intentionally intimate setting. The goal wasn't scale — it was depth.
The room was filled with people who care how models behave after the demo: latency, reliability, evals, iteration loops, and real user feedback. Conversations moved quickly from architecture to production tradeoffs to what open models unlock when builders truly understand them.

We were honored to hear from two outstanding speakers:

• Arya Talebzadeh (OpenAI) shared insights on the data flywheel — how frontier teams use user interactions, evaluations, and RLHF to continuously improve production systems.
• Vincent Koc (Comet, MIT) spoke about transitioning from closed models to open-source frontier models like MiniMax M2.1 , how to interpret benchmarks correctly, and how to extract real performance from agentic workflows.

We also walked through MiniMax M2.1 itself:
state-of-the-art open-source performance, a 10B active / 230B total MoE architecture , faster inference, local deployability, and why "open" is increasingly winning in real production environments.
The dinner that followed was exactly what we hoped for — builders trading notes, debating approaches, and forming new collaborations late into the night.

On the 10th, momentum was carried into the AI Agent & MCP Hackathon . Over two intense days, nearly 200 founders and engineers built AI agents, MCP systems, and scalable growth workflows — not prototypes for slides, but systems meant to run, iterate, and improve. Teams shipped fast, tested ideas through real outreach and traction experiments, and refined their approaches with feedback from experienced operators and judges.
Participants competed for $6,000 in prizes and guaranteed investor interviews, but the real value was in the execution: learning what works, what breaks, and how to iterate under real constraints.

This weekend captured exactly what we mean by our mission: Intelligence with Everyone .
Advanced capabilities shouldn't be locked behind walls — they should be open, usable, and shaped by the people who build with them every day.

To everyone who joined us — speakers, partners, hackers, founders, and engineers — thank you for bringing curiosity, rigor, and real experience into the room.

More to come. 🚀