Price War on Mid-Tier Models Slashes Anthropic and OpenAI Costs by 80 Percent

Prices plunge amid Asian competition

US AI labs OpenAI and Anthropic are sparking a massive price war. The primary source page was blocked during verification, so I am carefully relying on metadata and related signals rather than unverified detailed text. According to available indices, OpenAI slashed rates for its model GPT-5.6 Luna by a full 80%. Anthropic rolled out Claude Opus 5 at half the price of the previous Fable 5 generation.

This aggressive discounting comes exactly as European and US corporations have begun strictly capping internal AI spending. These companies are now actively testing and deploying considerably cheaper Chinese alternatives in production.

Budget pressure exposes per-token costs

Customers feel the financial burden very concretely. Both US companies are gradually moving a massive portion of their enterprise client base from flat-rate subscriptions to pure usage-based billing. Suddenly, management sees a hard link between the volume of processed data and the final monthly costs.

Such an environment paves the way for Chinese open-weight models. Free accessibility, local tweakability, and fractional operating costs are noticeably eroding the margins of Western leaders right before their planned trillion-dollar IPOs.

Computational effort emerges as the key variable

Headline prices for input and output tokens paint only part of the real-world deployment picture. A premium model seamlessly handles complex tasks on the first try using just a handful of data. Conversely, a cheap variant often relies on continuous query repetition and tedious developer corrections.

The focal point of budget optimization is rapidly shifting toward tuning the effort parameter. This allows applications to allocate extra compute power exclusively to complex queries where the massive investment in expensive tokens actually pays off.

Production integration remains the ultimate test

The current wave of discounts makes US mid-tier models a highly attractive option once again against Asian open-source competitors. The main battle will now be fought over integration with critical corporate systems.

If European and US banks or hospitals permanently hook their live data up to Moonshot, not even a zero-margin strategy will bring them back to US labs.

Lilith's verdict

Panic in the San Francisco accounting departments forced the market leaders to slash their prices by eighty percent. European and US corporations made it perfectly clear that they will happily route their data to China's DeepSeek if it saves them a million dollars by the end of the quarter.