On the evening of October 27, developers spotted DeepSeek-V5 model identifiers in the official DeepSeek Harness GitHub repository, including deepseek-v5, deepseek-v5-open, and deepseek-v5-next, listed alongside deepseek-v4-pro. The entries currently serve as framework-level placeholders rather than directly callable model interfaces.
Cognition released FrontierCode 2.0, a private coding benchmark with 132 real open-source repository tasks scored by whether maintainers would merge the changes. Claude Opus 5.5 leads at 46%, while DeepSeek V4.1 Flash is the top open-weights model at 28%.
DeepSeek's V4.1-Flash-1017, released on Oct 17, topped OpenRouter's weekly ranking (Oct 19-25) with 33.6 trillion tokens of usage, up 801% week over week. The updated version, based on the Sep 10 V4.1 Flash, optimizes code, agent, and visual tasks, while the older V4.1 Flash saw a 45% weekly decline.
DeepSeek is about to complete its second funding round, with the proposed amount raised to about 100 billion yuan at a pre-money valuation of 500 billion yuan. CATL, which invested about 5 billion yuan in the first round, plans to add another 10 billion yuan; Tencent and other existing shareholders will continue to participate, while Geely and SMIC's fund are also expected to join. If completed, DeepSeek's total funding within half a year will reach about 150 billion yuan.
Yuangu Tiantan, a new media account under China Media Group, revealed on October 22 that DeepSeek and MiniMax have completed registration procedures and finished open-sourcing and overseas licensing since the Ministry of Commerce and the Ministry of Science and Technology solicited public comments on revising the Catalogue of Technologies Prohibited or Restricted from Export on October 16.
Perfect World disclosed that two external special-purpose funds have completed fundraising, raising RMB 11.17 billion in total. Perfect World subscribed RMB 150 million, Century Huatong RMB 650 million, and Litong Electronics RMB 95 million. Lisi Xingshen completed its private fund filing on Sept. 15, serving as an external investment vehicle for DeepSeek's current financing round.
According to Bloomberg citing sources, DeepSeek is preparing for a STAR Market IPO while advancing a billion-yuan fundraising round, with plans to file its listing application by the end of 2026 and go public in 2027. Former Hillhouse Capital venture partner Yan Wentao has been appointed CFO.
In an interview with Yahoo Finance after GTC Berlin, Nvidia CEO Jensen Huang addressed investor concerns over AI infrastructure debt, saying Nvidia GPUs last seven to eight years and can serve as loan collateral, while citing DeepSeek's $38-per-hour GPU inference monetization price as evidence of strong demand.
DeepSeek is working closely with Huawei on AI chip design, with Huawei engineers embedded at DeepSeek to feed large-model training needs into chip design, as part of efforts to reduce reliance on Nvidia chips.
DeepSeek announced on October 20 that it has fully open-sourced its internal inference framework DeepInfer, which integrates high-performance operators and an optimized inference engine, supports distributed inference, speculative decoding, and structured output, and is compatible with NVIDIA and Huawei Ascend platforms. The company has used the framework to power high-throughput, low-latency inference for V4.1 Flash and also open-sourced the production-grade token codec component deepseek-recipe, aiming to reduce compute costs and energy consumption.