Building Abundant Intelligence
OpenAI CFO Sarah Friar outlines the company's economic framework for scaling AI infrastructure around a self-reinforcing cycle: falling inference costs expand addressable work, broader adoption funds R&D, and R&D improvements drive furth…
OpenAI CFO Sarah Friar outlines the company's economic framework for scaling AI infrastructure around a self-reinforcing cycle: falling inference costs expand addressable work, broader adoption funds R&D, and R&D improvements drive further efficiency gains. Recent pricing moves include an 80% price cut on GPT-5.6 Luna (now $0.20/M input tokens) and a 20% cut on GPT-5.6 Terra, with GPT-5.6 Sol delivering 2.5x speed in Fast mode at 2x cost. On the efficiency side, GPT-5.6 Sol helped reduce model-serving costs by 20% and improve token-generation efficiency by over 15%; context management improvements tripled ARC-AGI-3 benchmark scores while using 6x fewer output tokens. OpenAI now serves over 1 billion active users and 2 million businesses, with agentic workloads via Codex representing 99.8% of weekly output tokens, signaling a structural shift from query-based to task-completion AI use across enterprise functions.
- 01OpenAI CFO Sarah Friar outlines the company's economic framework for scaling AI infrastructure around a self-reinforcing cycle: falling inference costs expand addressable work, broader adoption funds R&D, and R&D improvements drive further efficiency gains.
- 02Recent pricing moves include an 80% price cut on GPT-5.6 Luna (now $0.20/M input tokens) and a 20% cut on GPT-5.6 Terra, with GPT-5.6 Sol delivering 2.5x speed in Fast mode at 2x cost.
- 03On the efficiency side, GPT-5.6 Sol helped reduce model-serving costs by 20% and improve token-generation efficiency by over 15%; context management improvements tripled ARC-AGI-3 benchmark scores while using 6x fewer output tokens.
- 04OpenAI now serves over 1 billion active users and 2 million businesses, with agentic workloads via Codex representing 99.8% of weekly output tokens, signaling a structural shift from query-based to task-completion AI use across enterprise functions.