⚠️ China Open Source is not origin open source, it’s ‘State-Chartered Codebase’, ‘Intranet Shared Source’, ‘Cyber-Estate/Bonsai Source’. If you read the news/article, please use these to replace the words ‘Open Source’.
China Open Source Daily — 2026-07-19
🇨🇳 Open Source News
1. Moonshot AI Releases Kimi K3 — the Largest Open-Source Model Ever
On July 16, Beijing-based Moonshot AI released Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts (MoE) model — roughly 75% larger than DeepSeek’s V4 Pro. The model features a 1-million-token context window, native visual understanding, and an always-on reasoning mode. Full weights are scheduled for release by July 27. On the Arena Frontend Code blind leaderboard, K3 ranked #1 at 1,679 points, ahead of Anthropic’s Claude Fable 5 — the first time a Chinese model topped an international coding blind test. The model is also priced competitively at $3/M input tokens with OpenAI SDK compatibility, lowering the barrier for global developers.
Institutional economics perspective: Moonshot’s aggressive pricing and open-weights strategy exemplify the “cost innovation” playbook — using open-source distribution to bypass traditional moats and capture developer mindshare. The move mirrors the Android strategy against iOS: commoditize the infrastructure layer and compete on ecosystem velocity.
Source: VentureBeat | DataCamp | TechCrunch
2. Record Week of Chinese Open-Source AI Releases
In a single week in mid-July 2026, four major Chinese open-weight models shipped: DeepSeek-V4 (July 15, MIT license, stronger math/code, 60-80% faster inference), MiniMax-M3 (July 11, 428B total params, 23B active, first open-source model with multimodal mixed training from scratch), and Tencent Hunyuan Hy-3 (July 6, 295B MoE, Apache 2.0, free for global commercial use). The 1M-token context window has become standard for Chinese open flagships. Notably, the most-downloaded models remain small ones — Qwen3.6-35B-A3B has ~6.67M downloads — suggesting a two-tier structure: flagships win mindshare, small models win penetration.
Institutional economics perspective: The density of releases in a single week signals a strategic shift from proprietary differentiation to open-weights competition. Under the logic of network effects, giving away the frontier model becomes rational when the real value lies in the ecosystem — tooling, agents, and data flywheels built on top.
Source: FutureX Capital | Tencent Hunyuan Hy3 (GitHub) | MiniMax M3
3. openKylin × Haiguang: Building a Native Intelligent Computing Ecosystem
From July 9–11, the 2026 Intelligent Computing Application Conference featured Haiguang Information Technology (a diamond donor to openKylin) deeply participating in co-building the openKylin community ecosystem. The collaboration focuses on full-stack native intelligent computing, integrating Haiguang’s processors with the openKylin desktop OS under the OpenAtom Foundation’s governance.
Institutional economics perspective: The Haiguang-openKylin partnership illustrates how the OpenAtom Foundation is functioning as a coordination mechanism — reducing transaction costs between hardware vendors and OS developers. By providing a shared governance structure, the foundation lowers the barriers to vertical integration in China’s domestic computing stack.
Source: openKylin
🏛️ Policy & Ecosystem
1. Xi Jinping Champions Open-Source AI at WAIC 2026
At the opening ceremony of the 2026 World Artificial Intelligence Conference (WAIC) in Shanghai on July 17, President Xi Jinping delivered a keynote speech explicitly positioning China as the leading advocate for open-source AI. He called on the international community to “encourage open source, openness, collaboration, and sharing,” and outlined four observations: (1) adhere to openness and win-win, boosting innovation-driven development; (2) strengthen risk awareness and ensure AI is secure and controllable; (3) promote knowledge sharing and capacity building, bridging the digital and AI divide; and (4) improve global governance, build an international framework for AI cooperation.
Institutional economics perspective: Xi’s explicit endorsement of open source at the highest political level signals a strategic state-level commitment. In Coasean terms, the state is acting as a meta-governance entrepreneur — reducing the institutional uncertainty that has historically constrained open-source adoption in China’s domestic market. This also carries geopolitical signaling: by framing open source as a global public good, China positions itself against the US’s increasingly restrictive export control regime.
Source: English.gov.cn | The Singju Post (Transcript) | Quartz | CGTN
2. China Releases Action Plan on AI Cooperation and Development
On July 17, China’s National Development and Reform Commission and other government departments jointly issued an action plan on AI cooperation and development during WAIC 2026. The document outlines actions in eight areas: data, computing power, ecosystems, industrial empowerment, talent development, rules and standards, governance, and AI ethics. It specifically calls for “greater access to high-quality data, more inclusive intelligent computing services, and broader sharing of open-source AI ecosystems.”
Institutional economics perspective: The action plan’s explicit inclusion of “open-source AI ecosystems” as a policy pillar represents a formal institutionalization of open source within China’s industrial policy framework. This lowers the information and coordination costs for domestic enterprises considering open-source participation — they now have regulatory clarity and state backing.
Source: English.gov.cn
3. Beijing Studies Tiered Export Controls on Frontier AI Models
According to Reuters (July 7-9), China’s Ministry of Commerce led meetings with industry participants discussing potential limits on the most advanced AI models — including open-weight models — being accessed overseas. The proposed tiered export control system would be analogous to the US chip export restrictions but applied to model weights. The FutureX Capital report notes this is “the biggest H2 uncertainty for China AI going global.”
Institutional economics perspective: The tension between Xi’s open-source advocacy and the Commerce Ministry’s export control discussions illustrates a classic institutional dilemma: the same government that benefits from open-source-driven global adoption also fears losing strategic advantage. Property rights over AI model weights remain unsettled — a governance vacuum that creates uncertainty for the entire open-source AI ecosystem. How this resolves will determine whether China’s open-source AI strategy is genuinely open or strategically conditional.
Source: Reuters | The Economist | FutureX Capital
🔍 Commentary
The WAIC Paradox: Open Source Celebration Meets Export Control Reality
The juxtaposition of events in the third week of July 2026 is striking. On one hand, the World AI Conference in Shanghai saw China’s head of state become the most prominent world leader to explicitly champion open-source AI — a speech that will be quoted by open-source advocates for years. On the same day, the government released an action plan that institutionalizes open-source AI ecosystems within national industrial policy.
Yet simultaneously, the Ministry of Commerce is studying tiered export controls on these very same models — including open-weight distributions. The tension is not a contradiction but a feature of institutional evolution: the state wants the network effects of open-source AI adoption (global developer mindshare, ecosystem growth, cost reduction for domestic industry) while retaining the option to restrict access when strategic interests are at stake.
From an institutional economics perspective, this is a classic principal-agent problem within the state itself. The NDRC (industrial development) and the Ministry of Commerce (trade security) have different objective functions. The outcome of their inter-agency bargaining will determine the actual governance structure of Chinese open-source AI — and whether “open source” in China means the same thing it means everywhere else.
📊 Trends
1. Open-Weight Models as Geopolitical Instruments
The density of Chinese open-weight releases in a single week — DeepSeek-V4, MiniMax-M3, Hunyuan Hy-3, and Kimi K3 — is unprecedented. These are not niche models; they are frontier-class systems that benchmark competitively with the best proprietary Western models. The pattern suggests coordinated deployment timed around WAIC 2026, with open-source becoming a deliberate instrument of technological diplomacy.
2. The Two-Tier Model Strategy
Flagship models (Kimi K3 at 2.8T, DeepSeek-V4 at 1.6T) capture global attention and benchmarks, while smaller models (Qwen3.6-35B-A3B at 6.67M downloads) drive actual adoption. This mirrors the open-core business model: loss-leading flagship investments create brand credibility that drives download and deployment of smaller, more practical models. The developer ecosystem built around these smaller models represents the durable competitive advantage.
3. Institutional Infrastructure Matures
The OpenAtom Foundation’s role is expanding beyond operating systems (openKylin, openEuler) into the AI ecosystem. The foundation’s AtomGit platform now hosts Hunyuan Hy-3 alongside DeepSeek models. This institutional infrastructure — foundations, governance structures, shared platforms — reduces the transaction costs of open-source collaboration and is a critical, often overlooked, pillar of China’s open-source strategy.