OpenAI and Anthropic are emphasizing lower-cost AI options as open-weight models attract more enterprise workloads.
Cost is becoming a bigger factor in AI model selection, particularly for high-volume and less demanding workloads. Several Chinese vendors now offer models that compete with leading US models on certain benchmarks while charging less per token.
OpenAI said it cut prices for its lower-cost GPT-5.6 Luna model by as much as 80%, depending on the type of token usage. Anthropic has also emphasized price in marketing Opus 5, which it says offers “frontier intelligence” at half the price of Mythos. Its Sonnet model is positioned as a balance between the company’s lightweight and highest-performing models.
According to data from SiliconData, via the Financial Times, there has been a noticeable shift in inference token spend away from closed models, which OpenAI and Anthropic primarily provide, since mid-July. While spending remains much higher on proprietary models than open-source alternatives, the gap has narrowed over the past month.
The pushback against increasingly high AI bills has come at a dangerous time for both OpenAI and Anthropic, which are planning IPOs for later this year or early next. Anthropic has reportedly been valued by private investors at as much as $2 trillion, which would make it the largest IPO ever. OpenAI is reportedly targeting a valuation above $1 trillion.
At the same time, the cost of securing more compute continues to rise, which will undoubtedly affect their ability to turn a profit in the near term. Anthropic was reportedly profitable in the second quarter of 2026, but is unlikely to repeat that feat over the next two quarters.
More must-read AI coverage
Chinese open-weight models compete on price
Several major US tech companies, which are typically among the biggest monthly spenders on AI models, have shifted some workloads to cheaper Chinese alternatives. DoorDash, Airbnb, and Coinbase are just a few of the companies that have confirmed using Chinese AI models.
While US models are often used during the development stages of an AI service, some companies are shifting production workloads or less intensive requests to cheaper Chinese models. Model-routing services such as OpenRouter make this far easier. Even so, security and compliance experts remain concerned about sending private or potentially sensitive data outside the country.
China’s AI model makers have had several major successes over the past few months. Moonshot AI’s Kimi K3 topped a frontend coding benchmark and has performed well across several other metrics. Alibaba, DeepSeek, and z.AI have also launched new models that have performed strongly against the current market leaders.
That momentum could soon face resistance, however, as both the Chinese and US governments have signaled concerns about US and European companies using Chinese AI technology. In the US, accusations of “industrial-scale theft” through model distillation have led several officials to warn of possible bans. In China, the government is weighing tougher export controls, particularly on frontier AI technology, as leverage against further US restrictions on Chinese technology.
For businesses using AI at scale, this shift means model selection is becoming less about choosing a single best-performing provider and more about balancing performance, cost, security, and regulatory risk across different workloads.
Read more: As Chinese developers compete more aggressively on price and performance, learn how their lower-cost AI models could introduce security and compliance trade-offs for businesses.
Other Articles
No Comment! Be the first one.