Jefferies: DeepSeekV4.1-Flash's price-performance ratio highlights Alibaba-W (09988), Tencent Holdings (00700), and MINIMAX-W (00100) “buy” ratings

The Zhitong Finance App learned that Jefferies released a research report saying that DeepSeek released DeepSeek V4.1-Flash, which is a cost-effective model; the bank rated Alibaba-W (09988), Tencent Holdings (00700), and MINIMAX-W (00100) as “purchases”, with target prices of HK$184, HK$750 and HK$533, respectively.

Jefferies pointed out that DeepSeek—V4.1-Flash has many breakthroughs, with total parameters reaching 552 billion (active parameters: 8 billion inputs, 16 billion outputs), uses a new Causal Encoder-Decoder (CED) architecture, and uses a new pre-training method for large-scale post-training; in third-party Agentic Benchmark evaluations, it outperformed Kimi-K3, Anthropic's Opus 5, and OpenAI's GPT5.6 Sol.

The bank notes that compared with previous models, HBM and SSD have been reduced to one-quarter and one-eighth, respectively. DeepSeek continues to use pricing during peak and off-peak periods. During peak periods, cache hit/input/output prices are 0.04 yuan, 2 yuan, and 8 yuan per million tokens, respectively, and 0.02 yuan, 1 yuan, and 4 yuan respectively during off-peak periods. Compared with the peak period, it is 87%, 78% and 70% lower than DS-v4-Pro; 98%, 90% and 92% lower than Kimi-K3, respectively; 98%, 75% and 71% lower than GLM-5.3, respectively. The bank believes that competition will intensify, and it is expected that peers will innovate in subsequent iterations.

The bank also indicated that after V4.1 Flash was released, the V4 Flash version reduced the peak cache hit/input/output prices by 60%, 33%, and 11%, respectively, to 0.04 yuan, 2 yuan, and 8 yuan per million tokens.