OpenAI and Anthropic Introduce New Models, Cut Costs as Cheaper Rivals Intensify Pressure

OpenAI and Anthropic are cutting the cost of their latest artificial intelligence models, signaling a shift in the frontier AI race from simply building more capable systems toward making advanced models cheaper and more efficient to operate.

OpenAI on Tuesday introduced two additional models in its GPT-6 family, GPT-6 Sol and GPT-6 Luna, cutting API prices by 50% compared with its promotional pricing for GPT-5.6.

Anthropic, meanwhile, unveiled Claude Opus 5.5, describing it as a more token-efficient version of its Opus 5 model that costs about 40% less to operate.

The simultaneous releases come as both companies face growing pressure from cheaper open-weight AI models, particularly from Chinese developers including Alibaba, Moonshot AI and DeepSeek.

The pricing moves also suggest that the economics of AI deployment are becoming as important as raw model performance. As businesses move from experimenting with AI to deploying models at scale, the cost of generating millions or billions of model responses can become a major factor in deciding which systems they use.

OpenAI said GPT-6 Sol is positioned below its Astra model and is designed for more complex workloads, including coding. GPT-6 Luna is aimed at "high-volume tasks" such as extracting information and summarizing documents. The two-tier approach allows OpenAI to match model capability and computational cost more closely with the task being performed. Customers do not necessarily need the most powerful model for routine workloads, while developers handling more complicated reasoning or coding tasks can pay for greater capability.

Anthropic is pursuing a similar efficiency strategy.

Claude Opus 5.5 is designed to use fewer tokens while maintaining the capabilities of its predecessor. Dianne Penn, Anthropic's head of product management, research and labs, said the company is working on making model reasoning and responses more efficient depending on the user's selected effort level.

"One of the things we're continuing to innovate on is how to make that thinking, how to make the answering more efficient, so it uses less tokens depending on your effort setting," Penn told CNBC.

That focus is deemed necessary because token consumption is closely linked to the cost of operating AI systems. More efficient models can potentially reduce the amount of computing infrastructure required to deliver the same volume of work.

The China Challenge

The pricing cuts also underpin the competitive pressure coming from open-weight models.

Alibaba, Moonshot AI and DeepSeek have been developing models that can compete with leading proprietary systems at substantially lower costs, putting pressure on OpenAI and Anthropic to justify the premium attached to their closed models.

For customers, the choice is increasingly becoming an economic calculation rather than simply a contest over benchmark performance. A company running millions of AI-powered customer interactions, coding tasks, or document-processing operations can generate a substantial difference in computing costs from even a modest reduction in the price of each request.

But that creates a difficult dynamic for frontier labs.

OpenAI and Anthropic are spending enormous amounts on computing infrastructure and model development, while cheaper competitors can put pressure on the prices they can charge customers. Lower API prices can stimulate demand and increase model usage, but they can also make it harder to recover the enormous cost of training and operating increasingly sophisticated systems.

The result is a race to improve the efficiency of both models and the infrastructure supporting them.

AI Spending Becomes The Next Battleground

The announcements also come at an unusual moment for the two companies. Anthropic CEO Dario Amodei recently called for an industry-wide slowdown in the development of advanced AI as concerns about model safety intensified. Former Anthropic researcher Jacob Coxon intensified that debate on September 8, saying he had left the company and warning on X that the industry's leading labs were "gambling with our lives."

OpenAI CEO Sam Altman and Tesla and SpaceX CEO Elon Musk also joined Amodei's call for greater attention to controlling the pace of AI development.

The latest product launches show that safety concerns have not removed the commercial pressure to keep improving AI systems. Instead, the companies are trying to make their models more economical while continuing to advance their capabilities.

That approach is expected to be increasingly adopted as AI spending expands across the technology industry. Against that backdrop, the next phase of the AI competition may be less about which company can produce the largest model and more about which company can deliver sufficient intelligence at the lowest cost.

The competitive threat from open-weight models has made the efficiency push more urgent. If cheaper models become good enough for a growing range of enterprise applications, the premium commanded by proprietary frontier systems could come under sustained pressure.

At the same time, the cost reductions could expand the overall AI market by making sophisticated models affordable for more developers and businesses. That creates a potentially important trade-off for the frontier labs: cheaper models could reduce revenue per unit of usage, but substantially greater usage could expand the market.

Industry analysts expect the outcome to depend on whether OpenAI and Anthropic can lower inference costs faster than competitors can close the capability gap.