DeepSeek has launched its V4.1-Flash AI model and is preparing for an IPO in Shanghai, aiming for a potential valuation of $75 billion.
Chinese AI company DeepSeek has introduced its latest model, DeepSeek-V4.1-Flash, which features native multimodality and a phased rollout of V4-Pro.
"The model is designed to enhance capabilities, expedite result delivery, increase throughput, and scale to larger models," the firm explained.
Enhanced Efficiency and Performance
DeepSeek claims that V4.1-Flash is the most compact model within its new architecture, utilizing a Mixture of Experts framework with 552 billion parameters. This model operates on a Causal Encoder-Decoder architecture, with 8 billion parameters active at input and 16 billion at output.
The company asserts that new pre-training techniques and extensive reinforcement post-training have improved the performance, cost-effectiveness, speed, and overall execution time compared to the previous DeepSeek-V4-Pro.
Notably, the V4.1-Flash model has a reduced requirement for KV-cache, needing only one-fourth of the High Bandwidth Memory (HBM) and one-eighth of the SSD storage compared to its predecessor, which is expected to lower costs in agent scenarios.
The model is accessible through the DeepSeek API, with previous models V4-Flash and V4-Flash-Vision-Exp being phased out. Requests to deepseek-v4-pro will automatically redirect to V4.1-Flash at the new pricing, effective from September 14, 2023.
IPO Preparations Underway
DeepSeek is preparing for its initial public offering (IPO) on the STAR Market of the Shanghai Stock Exchange, enlisting CITIC Securities as one of its underwriters. The company aims to kick off the IPO process by the end of 2026, although specific details about the offering are still pending.
The involvement of CITIC Securities suggests significant progress in DeepSeek's preparations, which usually involve a preliminary phase with brokerage firms prior to going public.
Potential Valuation of $75 Billion
In June, DeepSeek secured around $7.4 billion in funding, achieving a post-funding valuation exceeding $50 billion. Key investors include founder Liang Wenfeng, Tencent, and CATL. Reports in July indicated that a subsequent funding round could elevate DeepSeek's valuation to approximately 500 billion yuan, or $75 billion. The company plans to invest in computational infrastructure and model development to compete with other AI firms.
Liang Wenfeng aims to use the new capital to incentivize key researchers and engineers, as DeepSeek faces talent attrition to larger competitors like ByteDance and Xiaomi, which are also expanding their AI capabilities.
