Open Weights vs. Closed Weights: The End of the Performance Race?

For a long time, generative artificial intelligence was dominated by so-called “closed-weights” models: proprietary models, accessible only via API, developed by a few major players such as OpenAI and Anthropic. In contrast, “open-weights” models—which can be downloaded and deployed locally—have historically lagged significantly behind in terms of technology. But that era is coming to an end.

5 Key Takeaways

• Open-source models are catching up at an accelerating pace : the performance gap with proprietary models has narrowed from 6–7 months to less than 2 months, making the technological advantage of the AI giants less and less decisive.

• Performance is no longer the differentiating factor : For about 90% of enterprise use cases, the differences in quality among the best models have become virtually imperceptible for common tasks (analysis, generation, support, development).

• Four new criteria are emerging : price, data sovereignty, business specialization, and the ability to integrate with operational systems are becoming the true drivers of differentiation.

• The price war is structurally disadvantageous for OpenAI and Anthropic : Unlike Google or Alibaba, for which AI is not the primary source of revenue, these companies depend directly on monetizing their models.

• Value is shifting from the model to the system : The next area of competition will focus on autonomous agents and integration into business processes, rather than on the raw power of general-purpose models.

A Spectacular Technological Catch-Up

Just a few months ago, the release of Qwen 3 Coder Next had already marked a turning point: an open-weights model capable of achieving performance levels close to those of leading proprietary models such as Claude Sonnet or the latest generations of GPT in software development applications. Today, this trend is extending to general-purpose models.

With Qwen 3.5, Alibaba demonstrates that an open-weights model can now compete with frontier models across a wide range of benchmarks and professional use cases. The key point isn’t just raw performance—it’s the speed at which it’s catching up. Historically, the gap between open and closed models was estimated at about six to seven months. Today, it has fallen to less than two months. In other words, the technological advantage of proprietary models is becoming increasingly narrow and less and less strategic.

Performance is no longer the real issue

For about 90% of business use cases, the differences in quality among the top models become almost imperceptible:

• Document Generation
• Data Analysis
• Development Assistance
• Process Automation
• Business Support
• Chatbots

In this context, the nature of competition is changing. We are shifting from a mindset of “Which model is the smartest?” to one of “Which model is most relevant to my context?”

Four new criteria are now key factors

• The price
• Data Ownership and Sovereignty
• Business Specialization
• The Actual Use of Models and Their Integration into Operational Systems

The Issue of Price: An Asymmetrical Struggle

When it comes to costs, not all players are on equal footing. Companies like Google and Alibaba have a major structural advantage: AI is not their primary source of revenue. They can therefore offer very aggressive—or even subsidized—pricing to gain market share or strengthen their ecosystem. Conversely, for companies like OpenAI and Anthropic, the business model relies directly on monetizing their models. A price war will therefore be difficult for them to win in the long run.

Specialization: The Response from Closed-Weight Players

Faced with these constraints, leaders in the proprietary model space have adopted a new strategy: vertical specialization. We are already seeing the emergence of models specialized in law, medicine, finance, software development, and scientific research. This approach aims to regain a competitive advantage through domain expertise rather than raw computing power.

The Real Emerging Challenge: Usage

A fourth factor is now taking center stage: the use of models. Major technology companies have come to understand that value no longer lies solely in the model itself, but in its ability to be integrated into autonomous systems capable of taking action, reasoning through multiple steps, and automating complex tasks. Recent market trends point in this direction, as seen, for example, in OpenAI’s hiring of the creator of OpenClaw and Meta’s acquisition of ManusAI. These initiatives illustrate a shift toward agents capable of executing complete processes rather than simply generating text. The focus is thus gradually shifting from the model to the system.

DATASOLUTION’s Strategic Choice

At DATASOLUTION, our DATA & AI agency is convinced that the future of AI in business depends neither on the size of the models nor on a systematic reliance on major international platforms, but rather on the ability to design solutions tailored to the real-world context of organizations, such ase-commerce and web development.

Our approach focuses on smaller-scale models that are finely tuned to specific business use cases, trained or specialized using relevant company data, and then integrated into robust application architectures.

This strategy often delivers superior performance in real-world applications, while keeping infrastructure and operating costs under control.

It also offers strong guarantees regarding sovereignty, thanks to hosting options in France, Europe, or on-premises, thereby meeting the security and compliance requirements of many industries.

Finally, the use of more compact models helps reduce the energy footprint of AI solutions, which is an increasingly important issue in terms of environmental responsibility.

More broadly, we believe that value now lies in the combination of specialized models, domain-specific data, and intelligent systems capable of automating processes—rather than solely in the raw power of a general-purpose model.

The Next Battle in AI Won’t Be the One We Think It Will Be

The current phase marks a profound change in the industry.

The next competition will not be about who has the largest model or the highest benchmark, but about who delivers the most business value, keeps costs under control, ensures sovereignty, and integrates best with the information system.

Artificial intelligence is entering a phase of maturity, and in this phase, strategy matters more than raw power.

The fact that open-source models are catching up sends a strong signal: the technological advantage of the AI giants is no longer as sustainable as it once was. For businesses, this is excellent news. It opens the door to strategies that are more autonomous, more cost-effective, and more responsible. At DATASOLUTION, we’ve chosen a pragmatic, useful, and controlled approach to AI that’s aligned with the real challenges organizations face.