Update on AI progress. In the past 2 months, LLMs are still developing rapidly. Deepseek R1 has quickly fallen behind again.

This wave is mainly led by ChatGPT's comeback.

1. ChatGPT-4o's native multimodal capabilities have significantly improved the product experience for 2C users.

2. The O3 model's agent capabilities have been internalized, with a substantial increase in computational consumption during inference and overall performance. Moreover, the previous method of estimating market size by observing user-side input/output tokens has become obsolete.

Deepseek-r2 is under immense pressure. Market expectations are too high, and sentiment is overly optimistic, already pricing in significant progress and breakthroughs. Whether it can deliver remains a question.

$Alibaba(BABA.US) Qwen3 was released yesterday, roughly matching r1 in inference but with at least half the resource consumption (good news for $SMIC(00981.HK)). It also incorporates some features to catch up with o3 (mainly the agent part).

Meanwhile, other domestic large models like Xiaomi, Li Auto, and Doubao are, in my opinion, still in the stage of catching up to Deepseek R1. $Taiwan Semiconductor(TSM.US)  $NVIDIA(NVDA.US)