The biggest concern looming over Anthropic's IPO: models are getting cheaper to sell.

动察 Beating AI Flash: OpenAI CFO Sarah Friar said that after an 80% price cut on GPT-5.6 Luna at the end of July, usage surged 10-fold. Luna's current API price is only $0.20 per million input tokens and $1.20 per million output tokens.

By mid-August, Luna's token usage on OpenRouter had more than doubled the combined total of Anthropic Opus 5 and Sonnet 5. However, this isn't entirely due to the official price cut. OpenAI also offered an additional 50% discount on Luna and Terra through OpenRouter at the time, meaning Luna's effective price briefly dropped to roughly one-tenth of its original level.

Anthropic has begun lowering usage costs. The newly released Fable 5.1 keeps base input and output prices unchanged, but slashes cache read fees by 75%. Anthropic estimates that typical workloads could see bills drop by about 25%, while agent-heavy tasks could see reductions of up to 45%.

The price war coincides with Anthropic's preparations for an IPO. The company's annualized revenue run rate exceeded $65 billion by the end of July, with the prospectus expected to be made public this month. As model capabilities converge, if developers start switching models more frequently based on price, whether Anthropic can continue charging a premium for high-end models will become a calculation investors must revisit.