Summary of Key Points
Recently, there have been significant changes in the global large-model market: Chinese manufacturers have moved the "open-source" trend from the periphery to the center, forcing the entire industry to reprice its products. Leading companies such as Alibaba, DeepSeek, and Meta have been actively releasing open-source models. These models have transformed from being followers of closed-source technologies to challengers capable of competing with them. Not only have their performances greatly improved (for example, they can now independently complete complex tasks), but they have also found new ways to generate revenue (by charging fees from large users). As a result, closed-source models have been forced to lower their prices, while open-source models have become more expensive due to increased usage. This open-source competition is no longer just a small experiment within the tech community; it represents a major battle for global AI ecosystems, pricing power, and competitive landscapes.
I. Open-Source Models: From Follower to Challenger
In the past, open-source models were seen as inferior to closed-source ones, limited in performance and relegated to supporting roles. However, things have changed:
- Dramatic Performance Improvement: The programming ability score of DeepSeek V4 Pro has increased from 12.8 to 62.7; Alibaba's Qwen3.8-Max can independently program for 16 consecutive days and even operate a virtual e-commerce business; Meta's Muse Glimmer has been optimized for "agent" capabilities, acting like a small assistant that can plan, use tools, and debug on its own.
- Market Share Overturn: Chinese open-source models now account for 41% of global downloads, surpassing the United States for the first time, with cumulative downloads exceeding 10 billion times. Counterpoint suggests that if Western giants create "closed gardens" (preventing others from using their core technologies), developers and companies will turn to Chinese open-source models.
Open-source models are now on par with, or even challenging, closed-source models in terms of performance and have the potential to attract users and ecosystems, affecting the profits and valuations of closed-source solutions.
II. Open-Source Models Are Not Free: Revenue Comes from Large Users
Many people think that open-source means free access, but that's not the case. Modern open-source models combine "open accessibility with commercial licensing":
- Alibaba's New Approach: Alibaba plans to charge fees from large commercial users of Qwen3.8-Max—companies with annual revenue exceeding $20 million will need to sign agreements (the exact percentage is still under negotiation).
- A Precedent from Moonscape: The Kimi K3 model requires service providers with annual revenue over $20 million to pay up to 30% in fees.
This model allows small users to use the models for free or at a low cost while generating revenue from large users, balancing openness and commercial profitability.
III. Inverted Price War: Closed-Source Models Lower Prices, Open-Source Models Become More Expensive
Previously, closed-source models were more expensive, while open-source ones were cheaper. The situation has reversed:
- Closed-Source Models Forced to Reduce Prices: OpenAI has made its default model GPT-5.6 Luna available for free and offers unlimited chat services due to price competition among giants; they need to lower prices to retain users.
- Open-Source Models Become More Expensive Indirectly: This is not because manufacturers have increased prices, but rather because the increased usage has raised overall costs. For example, DeepSeek has raised its API fees and implemented "peak-valley pricing" (higher prices during peak times). The output price for DeepSeek V4-Pro has increased from 6 yuan to 27 yuan per million tokens.
Even with similar performance, open-source models are much cheaper than closed-source ones—DeepSeek V4-Pro costs about $0.87 per million tokens, compared to $6 for Grok 4.6, a seven-fold difference. This puts significant pressure on closed-source manufacturers, who risk losing both revenue and user bases.
IV. Strategic Shifts Among Global Players
The attitudes of leading companies towards open-source models have been evolving:
- Meta's U-Turn: Open-source Llama models became popular globally in 2023, but then Meta shifted back to using closed-source models due to the momentum of closed-source technologies. Now, Meta has returned to open-source, with Mark Zuckerberg writing a 6,000-word public letter stating its commitment to "making AI capabilities accessible to everyone."
- xAI's Hybrid Strategy: They use older open-source versions (like Grok-1) to attract users and build ecosystems, while more expensive closed-source versions (like Grok4.6) generate higher profits.
- The Struggles of U.S. Policy: Both Sundar Pichai and Mark Zuckerberg have called on the government not to restrict open-source development, noting that strict data and training regulations in the U.S. put domestic teams at a disadvantage compared to overseas ones (such as China). Banning foreign open-source models does not solve the problem.
V. Chinese Manufacturers Take the Lead in the Open-Source Race
Chinese manufacturers are at the forefront of this open-source revolution:
- Increase in Quantity and Quality: Domestic models like Kimi K3, Zhipu GLM-5.2, and MiniMax H3 have been released, with performance catching up to the world's top tier.
- Enhanced Market Influence: Their high download volumes have forced Western manufacturers to take notice of China's open-source ecosystem.
- Driving Industry Change: The low prices and high performance of Chinese open-source models have compelled closed-source manufacturers to lower their prices, reshaping the entire industry's pricing logic.
This open-source battle is essentially about Chinese manufacturers using "openness and cost-effectiveness" to break the monopoly of Western closed-source models and compete for control over AI ecosystems.
Conclusion
Open-source development is no longer just about idealism in the tech community; it has become a global competition to determine who can make AI more accessible and profitable. Chinese manufacturers are now at the center of this race, with the competition becoming increasingly fierce—focusing not only on performance but also on ecosystems, business models, and policy adaptability. For individuals, this means that using AI will become cheaper and more convenient in the future. For businesses, choosing open-source models can reduce costs and even allow them to contribute to the development of AI ecosystems.