China AI Intelligence: Key themes in H226E: model capability, monetization, and token ROI

Research evidence excerpt

China AI Intelligence: Key themes in H226E: model capability, monetization, and token ROI

Global Research

13 July 2026ab

China AI Intelligence Equities

ChinaKey themes in H226E: model capability,

monetization, and token ROI Internet Services

Wei Xiong

Analyst

wei.xiong@ubs.com

Three key themes +86-21-3866 8883

Investor interest in China AI models has increased after the listing of Zhipu and MiniMax Kenneth Fong

in early 2026, with market debates becoming more sophisticated across model iteration, Analyst

monetisation and competition. We set out three key themes to watch in H226E: 1) while kenneth-kc.fong@ubs.com

we expect continued improvement in model intelligence led by coding capability, we +852-3712 3890

flag early signs that SOTA models may defend their leadership, thanks to data Charles Chen

flywheel and improving user stickiness due to better productisation and the rising Analyst

adoption of AI in model R&D (e.g. recursive self-improvement); 2) we expect model- S1460524020001

layer monetisation to continue to ramp, as coding TAM expands beyond charles-za.chen@ubs.com

developers, while progress in multimodality may be overlooked; and 3) we expect an +86-21-3866 8907

increasing focus on token ROI, and the shift of enterprises and users from

tokenmaxxing to token optimisation should drive tiered demand and widen divergence

in the pricing power of state-of-the-art (SOTA) vs. other models, while Chinese models

may benefit from this trend given their structural cost efficiency (note).

Coding remains the clearest path for model intelligence improvement

We reiterate our view that coding is the core capability underpinning model intelligence

with a clear path to drive the capability frontier. We view AI coding as among the most