This move is a necessary macroeconomic measure amid astronomical operating expenses (reaching $34 billion last year) and a deficit of computing power. Instead of running a heavy AGI-like model for every prompt, the company is fragmenting the workloads. Sol will handle complex analytical tasks, Terra will strike a balance between speed and cost, and Luna will cover the needs of fast edge computing. Currently, access is restricted to vetted partners only. Ahead of its upcoming IPO, OpenAI must demonstrate to Wall Street that it not only knows how to burn cash on data centers but also how to flexibly manage the unit economics of its APIs.
Source: OpenAI / Reuters
LLMOpenAIGPT-5.6B2BAPI