Why Build Another AI Model

Agnes AI: Southeast Asia's Answer to ChatGPT (And 20x Cheaper) - Bruce Yang

Topic 2 of 16

Published Feb 2, 2026 · Clip 6:16–9:51 (3:35)

Bruce explains why Agnes built its own model despite thousands of existing ones. He describes the narrowing gap between closed-source SOTA models and open-source ones (DeepSeek, Qwen, Llama), and how Agnes's research team, user data from ~4 million registered users and ~300K DAU, and reinforcement learning allow them to close that gap. He clarifies the model is not yet open source, but a paper is published and a bigger open-source release is coming, with smaller 7B/30B models already reaching SOTA on Southeast Asian benchmarks.