Daily Highlights
- DeepSeek's new funding round grows to about 100 billion yuan
- Nvidia in talks to buy open-model startup Reflection AI
- DeepMind's AlphaProof Nexus cracks 9 open math problems
- StepFun's Step 5 Preview lands in Tencent WorkBuddy
- Modusi Intelligence raises over 1 billion yuan in Pre-A round
TODAY'S HEADLINES
[DeepSeek's New Funding Round Grows to About 100 Billion Yuan at Unchanged Valuation] On Oct 13, Caijing reported that DeepSeek's latest funding round is near completion, with the target raised from a planned 50 billion yuan to about 100 billion yuan in under two weeks, while the pre-money valuation holds at roughly 500 billion yuan. New investors are mainly corporate VCs, with Tencent and CATL among existing backers adding more; the company has also hired a CFO as it prepares for an IPO. (Source: IT Home/Caijing)
[Nvidia in Talks to Acquire Open-Model Startup Reflection AI at ~$25B Valuation] The Financial Times reported on Oct 10 that Nvidia is in early talks with Reflection AI over a full acquisition, a larger stake, or an acqui-hire that licenses technology and hires the team. Founded by ex-DeepMind researchers, the startup just released its first open-weight model Beam and is raising at a $25 billion pre-money valuation; Nvidia has already invested $800 million. (Source: Financial Times)
[DeepMind's AlphaProof Nexus Solves 9 Long-Open Math Problems, Published in Science] Google DeepMind's AlphaProof Nexus has a paper out in Science: the system solved 9 of 353 open Erdos problems, two of which had stood for 56 years, and proved 44 OEIS conjectures. Candidate proofs generated by LLMs are iterated via Lean compiler feedback, and all proofs are open-sourced on GitHub. (Source: The Decoder)
NEW PRODUCTS & TECH
[StepFun's Step 5 Preview Integrates into Tencent WorkBuddy, Free for Two Weeks] On Oct 12, Tencent's all-in-one AI workspace WorkBuddy integrated StepFun's latest flagship model Step 5 Preview, exclusively free for two weeks. Built for reasoning and agentic tasks, it handles deep research, information retrieval, data analysis and document generation, with sustained execution of multi-step tasks. (Source: National Business Daily)
[Microsoft Ships Decision-Scoring Model Microsoft-Decision-1 Built on Qwen3.5-9B] Microsoft released Microsoft-Decision-1 in October, fine-tuned on Alibaba's open-source Qwen3.5-9B. It posts an 83.5% average accuracy across 36 blind-test benchmarks (~150K questions), an 85 ms P50 latency—about 35x faster than GPT-6 Sol—and costs $0.042 per million input tokens. (Source: TMTPost)
IPO & FUNDINGS
[Modusi Intelligence Closes Over 1 Billion Yuan Pre-A Round Led by Huaxing Fund] On Oct 12, multimodal model startup Modusi Intelligence (MoSi) announced a Pre-A round of over 1 billion yuan led by Huaxing Fund, with Guokai Sci-Tech Innovation and Shanghai S&T Innovation Fund participating. Funds go to omni-modal real-time interaction R&D and global expansion; founder Qiu Xipeng is a distinguished professor at Fudan University. (Source: National Business Daily)
[DeepSeek to Lead Seed Round in Hu Han's Winverse AI at ~$400M Valuation] On Oct 13, Securities Times reported DeepSeek plans to lead a seed round of tens of millions of dollars in robotics startup Winverse AI, with Meituan Longzhu participating. Founded months ago by Hu Han, formerly head of Tencent Hunyuan multimodal models, the company builds embodied-intelligence foundation models. (Source: Securities Times)
OPEN SOURCE & RESEARCH
[AlphaProof Nexus Proofs Open-Sourced: Formal Verification Becomes the Paradigm] DeepMind open-sourced all Lean formal proofs from AlphaProof Nexus on GitHub; the system pairs LLM generation with compiler feedback across multiple agents. The paper shows base agents without the AlphaProof tool solved none of the Erdos problems, making the generate-verify loop a template for AI-driven research. (Source: GitHub/Science)
COMPANY & PEOPLE
[Xiaomi's LLM Team Settled: Head Luo Fuli Promoted to Top Grade 22] According to LatePost, Xiaomi's Sept 22 internal promotion list saw AI model team head Luo Fuli promoted, with people close to the company saying she reached grade 22, the top of Xiaomi's ladder, where further advances come mainly as role changes—signaling continued commitment to its model business. (Source: IT Home)
[DeepSeek Hires Ex-Hillhouse Investor Yan Wentao as CFO, IPO Prep Speeds Up] Chinese media report DeepSeek has hired Yan Wentao, formerly a partner at Hillhouse Venture Capital, as CFO. Yan previously invested in MiniMax, Zhipu AI and ByteDance; the hire is widely seen as pre-IPO positioning, with the company reportedly having engaged CITIC Securities for a STAR Market listing. (Source: East Money)
Powered by AOYII.com
