Western open-weight campaign could give enterprises more control over AI

China's open-model lead is not permanent

Chinese companies have established a leading position in open-weight AI, with models from DeepSeek, Alibaba, Z.ai and Moonshot gaining traction for their capabilities and relatively low operating costs.

Counterpoint Research has identified Chinese models, including Kimi and GLM, among the top-performing open-weight systems in the market.

Prince Kohli, president and CEO of Sauce Labs, a cloud-based software testing platform that has worked with open-weight vendors, said Chinese developers' early lead has been driven in part by necessity. Resource constraints pushed them to prioritize cost and hardware efficiency earlier, while Western AI companies concentrated more resources on frontier-model performance.

"That has given Chinese developers an early start and therefore a point-in-time advantage in price-performance and open-weight experimentation," Kohli said.

Kohli expects the gap between Western and Chinese models to narrow as more Western engineering resources shift toward optimization, noting that "nothing is fundamentally preventing Western developers from making the same cost and hardware improvements that gave Chinese models an early advantage."

While the models still need independent testing, they don't have to lead every benchmark to be useful to enterprises. "They need to be capable enough for specific workloads and offer a viable alternative to proprietary models," Kohli said.

For enterprises, the significance of that shift goes beyond model performance. It's also about what they can do with a model once they deploy it.

Control matters more than cost

For enterprises, the appeal of open source AI is often greater autonomy rather than lower cost. With an open-weight model, a company can run the system in its own environment, keep sensitive data within that environment and decide when to upgrade to a new model version. It can also fine-tune the model using proprietary data.

Kohli said self-deployment is especially important because it gives enterprises greater oversight of their data, which can be critical in defense, government and highly regulated industries.

"Fine-tuning also allows enterprises to adapt models using private, domain-specific data, creating AI capabilities they can control as intellectual property," he said. Organizations can also set their own model update and deprecation policies, reducing the risk of sudden changes in performance or quality.

Cost can still matter at high and steady volumes, but once enterprises factor in GPUs, staffing and evaluation, open weights aren't necessarily cheaper for smaller workloads, Pattanaik said.

The bigger advantage is how much of the AI stack enterprises can manage themselves, and that also makes portability important. "Keeping applications, prompts and AI logic separate from any one model provider makes it easier to switch models as needs change," Pattanaik said.

Enterprises can also use different models for different jobs: a smaller open model for high-volume, routine workloads and a proprietary frontier model for tasks requiring more advanced reasoning, he said.

But greater control also means greater responsibility. Open weights -- as opposed to the public training data offered by fully open source systems --  don't necessarily provide transparency into how a model was trained or what data was used to create it. A company can control where a model runs without being able to fully audit what's inside the model.

Licensing also varies, making the terms governing commercial use, modification and redistribution an important part of enterprise evaluation. Organizations running their own models also take on more responsibility for security, performance evaluation, updates and governance.

"You can't audit a probability, so the controls have to sit outside the model whether it's open or closed," Pattanaik said.

Open weights, therefore, aren't a shortcut around AI governance. Enterprises still need controls on the model for security, evaluation, updates and other governance processes. 

The real test is enterprise adoption

Reflection AI and Mistral don't need to permanently overtake Chinese developers for enterprises to benefit from a stronger Western open-weight market. Their success will depend on whether enterprises find the models capable, reliable and trustworthy enough for real workloads.

That could change how enterprises approach model procurement. Rather than choosing between proprietary Western models and Chinese open weights, organizations could use a broader mix of models based on the workload, regulatory requirements and level of control they need.

As Western and Chinese models converge in capability, Kohli expects other factors to play a larger role in adoption, including quality, reliability, cost and trust. Once token costs are sufficiently close, he said, enterprises will have more reason to evaluate how consistently models perform and whether they can rely on the vendor's policies and data provenance.

"Enterprises need predictable results, strong performance on complex use cases and the flexibility to maintain model choice rather than locking themselves into one provider," Kohli said. "In addition, they must trust the data provenance and policies of the vendor."

Kinza Yasar covers AI and emerging technology for TechTarget, with a focus on ethics, enterprise adoption, governance and business strategy. Before moving into journalism, she worked in IT and network support roles, giving her a systems-level perspective on how enterprise technologies are built, deployed and managed.