OpenAI Slashes Prices by 80 Percent: Chinese Competition Reshapes the AI Market
OpenAI and Anthropic are drastically cutting the prices of their AI models. The main driver behind these double-digit percentage discounts is aggressive Chinese competition, offering high performance at a fraction of the cost, which is changing the rules of the game for companies worldwide.
A New Era of AI Price Wars: OpenAI and Anthropic React to Pressure from the East
U.S. technology leaders OpenAI and Anthropic are entering a new phase of competition where aggressive pricing strategies take center stage. The reason for this shift is not just the accelerating pace of model development itself, but primarily the growing ambitions of Chinese competitors. They are launching increasingly capable systems into the market at a fraction of the price of Western alternatives.
For the corporate sector and developers who utilize generative AI on a massive scale, the cost of millions of processed tokens is becoming a critical budget item. Cost efficiency is thus starting to compete even with the raw performance of the models themselves.
OpenAI and Anthropic: Drastic Price Cuts for the Mid-Range
At the end of July, OpenAI sent a clear signal to the market with significant price reductions for its key models. Users of the GPT-5.6 Luna model benefit the most, seeing a staggering 80% price drop.
Current OpenAI Price Movements:
- GPT-5.6 Luna: Price reduction to $0.20 per million input tokens and $1.20 per million output tokens.
- GPT-5.6 Terra: Price drop of 20%.
- GPT-5.6 Sol: The price of the most powerful model remains unchanged, as the company seeks to protect margins for its flagship product.
Anthropic has adopted a similar strategy. The Claude Opus 5 model is now available for $5 (input) and $25 (output) per million tokens, representing half the price of the older Fable 5 model. Furthermore, in August, the company backed away from a planned price increase for the Sonnet 5 model, whose rate remains stable at $2 for input and $10 for output.
According to data from the Financial Times and the Silicon Data index, the actual inference costs for customers at leading U.S. labs have fallen by nearly a quarter since mid-July.
The Chinese Offensive: When Performance is Cheaper Than Coffee
The greatest pressure on pricing is currently coming from China. Companies like DeepSeek are demonstrating that top-tier models can be operated at extremely low costs. For the V4-Flash model, the price was set at just $0.14 per million input tokens and $0.28 per million output tokens.
The analytical firm Artificial Analysis estimated that the average cost for a single benchmark task on this model is a mere $0.03, while maintaining a very high level of performance.
"The price per token alone is ceasing to be the only metric. Companies are increasingly testing multiple models based on specific use cases and deploying the most powerful systems only where their higher price truly makes sense."
New Strategies for Corporate Customers:
- Hybrid deployment: A combination of expensive models for complex reasoning and low-cost models for routine tasks.
- Focus on efficiency: A more expensive model may be more cost-effective if it completes the task in fewer steps (tokens).
- Testing alternatives: Giants like DoorDash or Airbnb are already actively testing cheaper Chinese models for their operations.
The Future of the Market and Pressure on Margins
The price war comes at a time when both OpenAI and Anthropic are investing billions of dollars into infrastructure and computing power. They face a difficult task: they must prove to investors that they can turn these enormous costs into a long-term sustainable business model.
While the price drop is welcome news for end-users and companies, accelerating the real-world adoption of AI, for technology labs it means further pressure on margins and the necessity of constant operational cost optimization.