Chinese AI company DeepSeek has released the DeepSeek-V4.1-Flash model featuring native multimodality and announced the phased rollout of V4-Pro.
“Designed to enhance capabilities, accelerate output, increase throughput, and scale to larger models,” the statement said.
Compactness and Performance
The company describes V4.1-Flash as the most compact model in its new architectural family. It is a mixture of experts neural network with 552 billion parameters using a causal encoder-decoder architecture: 8 billion parameters are active on input and 16 billion on output.
DeepSeek claims that through new pre-training methods and more extensive reinforcement learning, the model surpasses DeepSeek-V4-Pro in performance, cost, speed, and overall execution time.
Separately, DeepSeek announced reduced requirements for KV-cache. V4.1-Flash needs only 1/4 of the HBM and 1/8 of the SSD storage compared to the previous generation. The company believes this will reduce costs in agent scenarios.
The model is now available in the DeepSeek API. DeepSeek has decommissioned V4-Flash and V4-Flash-Vision-Exp—compatible aliases deepseek-v4-flash and deepseek-v4-flash-vision-exp, temporarily redirecting requests to V4.1-Flash.
Starting at 04:00 UTC on September 14, all requests to deepseek-v4-pro will automatically route to V4.1-Flash under the new model’s rates. This mode will remain until the launch of V4.1-Pro. The updated pricing for V4.1-Flash took effect at 04:00 UTC on September 10.
DeepSeek also announced plans to collaborate with the open-source community to support V4.1-Flash inference and explore additional deployment options.
Preparing for IPO
DeepSeek has hired CITIC Securities to prepare for an initial public offering (IPO) on the STAR Market of the Shanghai Stock Exchange, according to Reuters and SCMP citing their sources. China’s largest investment bank will be one of four underwriters.
The company aims to initiate the IPO process by the end of 2026. The exact timing, size, and valuation of DeepSeek are yet to be determined.
Chinese firms typically undergo a preparatory procedure with broking firms before going public. The involvement of CITIC Securities indicates that DeepSeek has advanced further in its IPO preparations.
DeepSeek Could Be Valued at $75 Billion
In June, DeepSeek raised approximately $7.4 billion at a post-round valuation exceeding $50 billion. Major investors included company founder Liang Wenfeng, Tencent, and battery manufacturer CATL.
In July, Reuters reported that a new funding round could value DeepSeek at around 500 billion yuan, or about $75 billion. The company plans to allocate additional funds to computing infrastructure, model development, and employee retention amid competition with other Chinese and American AI developers.
According to the agency, Liang Wenfeng also aims to use the capital to enhance motivation for key researchers and engineers. DeepSeek has already faced the departure of specialists to larger competitors, including ByteDance and Xiaomi.
In July, media reported that the TikTok operator is shifting its focus to artificial intelligence and actively expanding its related infrastructure.
