Foreign media report that OpenAI is considering lowering prices for its models aimed at developers and enterprises to preempt a similar move by Anthropic. Both companies have secretly filed IPO applications this month, but neither has yet achieved stable profitability, making pricing competition a growing focus.
The article cites The Wall Street Journal, stating that the discussions are still evolving. Sam Altman recently also said that OpenAI will seek more ways to help users get more value for less money. However, the report notes that OpenAI’s adjusted operating profit margin for the first quarter of 2026 was -122%.
Anthropic is growing faster
The article suggests that OpenAI's consideration of price reductions at this time is related to Anthropic's recent advances over the past few months. Previously disclosed industry data shows that OpenAI's share of global generative AI web traffic has declined from 77.6% in May 2025 to 53.7% in April 2026.
Meanwhile, among companies tracked by the Ramp AI Index, the number of companies paying for Anthropic has surpassed those paying for OpenAI for the first time. The article states that Anthropic’s annualized revenue rose from $9 billion at the end of 2025 to $47 billion in May 2026, driven primarily by Claude Code.
Enterprise clients already have a low-cost alternative.
The core argument of the article is that even if OpenAI and Anthropic engage in a price war, cheaper alternatives are already available in the market. An increasing number of open-source inference providers are now offering inference capabilities for DeepSeek V4 and other Chinese models at prices significantly lower than proprietary models.
- Open-source inference service providers can fulfill enterprise API requirements.
- Chinese models are significantly cheaper than closed-source products.
- Pay-as-you-go pricing makes it easier for businesses to compare costs.
The article mentions that these platforms offer models including DeepSeek, GLM, MiMo, Kimi, and Minimax. In programming and other benchmark tests, these models can compete with Claude Opus, yet their prices are approximately one-thirteenth of the latter's.
Open source continues to press down the price floor
The article argues that this also makes the path represented by DeepSeek more compelling. The logic is not merely that the models are cheaper, but that open source itself is continuously lowering the lower bound of inference pricing. Inference service providers do not need to bear the core costs of model development; they only need to provide computing power and deployment services, allowing overall pricing to continue to decline.
Under this framework, the challenge for OpenAI and Anthropic is not just about competing for customers, but about maintaining profitability. The article notes that if China’s leading AI labs continue to adhere to an open-source approach, it will become even harder for closed-source model providers to restore high profit margins.