JPMorgan has released a research report on China's large language model industry, noting that as monetization paths converge on enterprise workflows, API consumption, code, and agents, investors will increasingly focus on a model's actual capabilities, task completion rates, and pricing power. In a context where AI adoption demand still outpaces inference computing supply, a developer's rapid retreat from premium pricing reflects a lack of market recognition for its model's improved capabilities. The firm expresses a relatively positive view on KNOWLEDGE ATLAS (02513) within the sector but has downgraded its investment rating for MINIMAX-WP (00100).

The bank reaffirmed its "overweight" rating on KNOWLEDGE ATLAS. It favors the "price maker" behavior demonstrated by the company, which doubled its API prices this year while maintaining continuous business volume growth, fully validating the market's recognition of its large model's value. JPMorgan has raised its revenue forecast for KNOWLEDGE ATLAS for fiscal years 2026 to 2030 by 26% to 42%, lowered its adjusted net loss forecast, and significantly increased its target price from HK$950 to HK$1,400.

Conversely, JPMorgan downgraded its investment rating for MINIMAX-WP from "overweight" to "neutral" and substantially cut its target price from HK$1,100 to HK$400. This is primarily due to MiniMax pricing its newly launched M3 model at twice the rate of the previous-generation M2.7 on June 8, only to permanently slash the price by 50% just one week later, bringing it back to a level comparable to M2.7. This action suggests a decrease in the visibility of its leading position in large models. To reflect the negative impact of the price cut on margins, the firm has reduced its revenue forecast for MINIMAX-WP for fiscal years 2028 to 2030 by 5% to 21% and widened its adjusted net loss forecast for fiscal years 2026 to 2028.