最猛清华副教授 7个月融了27亿

Bottom line: This article's key value is that Chinese AI lab Naive.AI raised about RMB 2.7 billion in seven months and released an open-source large model aimed at coding and AI R&D.

1. Key facts: Founded by Tsinghua associate professor Dai Jifeng, Naive.AI was established in February this year. Three funding rounds total about $400 million, with a post-money valuation of roughly $1.42 billion. Investors include Tencent, Sequoia China, IDG Capital, and Matrix Partners China.

2. Product details: The first open-source model, Naive-N0.5-Flash, has 309B total parameters, native support for a 1M-token context, and AI-optimized inference speeds up to 2,000 tokens per second. During development, the AI model wrote code, ran experiments, monitored progress, analyzed results, and iterated continuously.

3. Practical implication: This is a NeoLab-style company, meaning it secured substantial funding based on technical direction and founding team strength before a formal product launch. Domestic NeoLabs are more pragmatic, with clear product forms and commercialization front-loaded. Ordinary readers can focus on whether this practicality is actually delivered.

Bottom line: For brands, this article lays out a positioning, R&D, and commercialization playbook for an AI product, centered on a practicality-first approach.

1. Product positioning and cost-effectiveness: Naive.AI is not competing on general-purpose large models; it focuses on coding and AI R&D scenarios. The model is open source, supports a 1M-token context, and reaches up to 2,000 tokens per second. Previously, Dai Jifeng's team achieved 56.1 points on benchmarks with MiroThinker 1.5 at 30B parameters, compared with 60.2 points from Kimi-K2-Thinking at 1T parameters. This is a high cost-performance route.

2. R&D and iteration: Naive-N0.5-Flash was co-developed by researchers and AI, with AI writing code, running experiments, monitoring progress, and analyzing results. This suggests brands can use AI in R&D to shorten iteration cycles. The architecture is based on Xiaomi's open-source MiMo-V2.5 and uses DeepSeek sparse attention, showing that efficiency optimization on an open-source foundation can create competitiveness.

3. Brand and market observations: Domestic NeoLabs differ from Silicon Valley's pure technology belief model by emphasizing clear product form and front-loaded commercialization. The signal for brands is that capital is willing to pay a valuation premium for people plus frontier direction, but Chinese-style projects ultimately need scenario traction and practicality to prove brand value.

Bottom line: For sellers, the useful content here is the latest shift in AI investment and entrepreneurship, plus market opportunities and risks worth watching.

1. Growth market and opportunity: AI coding and AI R&D are clear growth areas. Naive-N0.5-Flash is purpose-built for coding and AI R&D, open source, with native 1M-token context support and inference speed up to 2,000 tokens per second; it may drive demand for AI tools, compute resources, and developer services. The rise of domestic NeoLabs with more front-loaded product and commercialization plans means related application and service ecosystems may mature faster.

2. Latest business models and cooperation: Naive.AI and Lin Junyang's p7k represent the raise-money-first-and-then-polish-the-product model. Naive.AI is backed by Tencent, Sequoia China, IDG Capital, Matrix Partners China, and other top institutions; Tencent also invested in related projects. Sellers can track the ecosystem and channel resources these institutions bring to AI products.

3. Risk and response: The article notes that NeoLabs can raise huge funding before any product is visible, essentially betting on founders and direction. Chinese NeoLab valuations are significantly lower than Silicon Valley peers, and funding amounts also diverge. For people selling or evaluating AI products, valuation alone is not enough; the key question is whether practicality can be realized and commercialization delivered.

Bottom line: For factories, the most valuable takeaway is that AI can now participate deeply in R&D processes, and enterprises can bring such capabilities into product design, production optimization, and digitalization.

1. Product and R&D implications: Naive-N0.5-Flash was developed with AI writing code, running experiments, monitoring progress, analyzing results, and iterating. This shows AI is not just a tool but an executor inside the R&D workflow. Factories pursuing smart manufacturing or digital transformation can watch its use in automated programming, system maintenance, and data analysis.

2. Technology selection clues: The model is based on Xiaomi's open-source MiMo-V2.5, replaces global attention with DeepSeek sparse attention, supports a native 1M-token context, and delivers inference speeds up to 2,000 tokens per second. When evaluating AI suppliers, factories can prioritize solutions optimized for long context, high throughput, and low-cost inference to handle large volumes of equipment data and process documents.

3. Business opportunity and cooperation: Domestic NeoLabs emphasize scenario landing and product practicality. Open-source large models mean manufacturers can do secondary development on top of base capabilities. But these companies are seeing fast valuation growth while product commercialization is still early; real-world deployment still requires validation of stability and actual benefits.

Bottom line: For service providers, this article can be distilled as follows: AI large models are moving from general-purpose conversation toward vertical scenarios such as coding and AI R&D, bringing new opportunities in tech stacks and business models.

1. Industry trend: NeoLabs are becoming a new force. Since 2024, more than 40 NeoLabs have cumulatively raised over $40 billion. In China, Naive.AI and p7k both received funding from top institutions before their formal products, but valuations are weaker than Silicon Valley projects. Domestic projects are more pragmatic, with product form and commercialization front-loaded. Service providers can treat this as a demand signal: enterprises want AI that solves concrete problems, not pure technology belief.

2. New technology and solutions: Naive-N0.5-Flash has 309B total parameters, native 1M-token context support, and AI-optimized inference up to 2,000 tokens per second, built specifically for coding and AI R&D. It is based on open-source Xiaomi MiMo-V2.5 and uses DeepSeek sparse attention, showing that an open-source base plus sparse attention is an efficient model route. When delivering AI applications, service providers can use similar models for long codebases, large documents, and R&D automation workflows.

3. Client pain points and implementation: AI models that write code, run experiments, monitor progress, and analyze results directly target pain points in R&D labor costs and iteration speed. The article also warns that Chinese NeoLab valuations lag Silicon Valley significantly, and practicality will be the metric tested. Service providers should focus on quantifiable outcomes and scenario landing, avoiding decisions based on high valuations alone.

Bottom line: For platform businesses, the core signal is that leading platforms are using investments to stake out new AI labs, and the ecosystem value of AI startups matters more than short-term products.

1. Latest platform play: Tencent appears in the investor lists for Naive.AI and p7k, joining Sequoia China, IDG Capital, Matrix Partners China, and others in betting on founders. This shows platform companies treat investing in people plus direction as an entry point into the next-generation AI ecosystem. Platforms can learn: instead of waiting for mature products, build an investment or partnership portfolio around key people and technical directions.

2. Platform demand and opportunities: Companies like Naive.AI target coding and AI R&D, and open-source models can attract developers, technical teams, and enterprise clients, potentially driving demand for model services, compute, toolchains, and cloud resources. Platforms offering open model downloads, inference services, or industry app marketplaces can use such marquee projects to strengthen merchant acquisition and operations.

3. Trend and risk mitigation: The article stresses that domestic NeoLabs are more pragmatic, with clear product forms and front-loaded commercialization, but valuations remain far lower than Silicon Valley peers. AI startups ultimately need practicality to prove their value. When platforms cooperate with such companies, they should track product rollout pace, user feedback, and commercialization metrics instead of committing resources based on valuation alone.

Bottom line: This article offers a new sample for studying AI industry organization and financing models: Chinese NeoLabs represented by Dai Jifeng's Naive.AI and Lin Junyang's p7k are forming a pragmatist entrepreneurial route distinct from Silicon Valley.

1. New industry dynamics: NeoLabs are mostly founded by alumni of top AI institutions such as OpenAI, DeepMind, and Anthropic. Since 2024, more than 40 have cumulatively raised over $40 billion. In the Chinese version, Naive.AI raised $400 million across three rounds in seven months at a $1.42 billion valuation, while p7k's first round was valued at about $2 billion. This reflects capital shifting from betting on products to betting on people and technical direction.

2. Talent and route comparison: On talent supply, Microsoft Research Asia and SenseTime connect the history of Chinese AI entrepreneurship. Companies such as Megvii, SenseTime, MiniMax, VAST, and Momenta were founded by related talent, and Dai Jifeng comes from that lineage. The China-US difference is clear: Silicon Valley emphasizes foundation model innovation and parameter scale, while China's Naive.AI released an open-source model after just over seven months; p7k's product form is an agent, focusing more on scenario landing and commercialization.

3. Business model and implications: Dai Jifeng continues a lightweight model route. Naive-N0.5-Flash is based on open-source Xiaomi MiMo-V2.5, replaces global attention with DeepSeek sparse attention, and uses AI models directly in R&D. This represents an open-source base plus efficiency optimization plus AI-collaborative R&D model. Researchers can further examine what high-speed fundraising projects with valuations ahead of products mean for innovation incentives, market screening, and risk control.