← All stories
AI & Tech

ChatGPT Cost Per Token Dropped 97% in Two Years From Version 4 to 5.4

All-In Podcast · OpenAI CFO Sarah Friar on IPO, AI Rivalries, New Device, and Spending $100B+ on Compute · June 2, 2026
ChatGPT Cost Per Token Dropped 97% in Two Years From Version 4 to 5.4
All-In Podcast
All-In Podcast
OpenAI CFO Sarah Friar on IPO, AI Rivalries, New Device, and Spending $100B+ on Compute
"From 4 to 5.4, the deprecation cost was 97%. That happened in like 2 years. That's incredible. Even our newest model, if you look at 5.5 that we just released, we actually raised prices on 5.5 2x. But if you look at what the cost to the customer is, they're probably still getting a break of about 20% to 30% cost reduction per token."
Friar disclosed that OpenAI achieved a 97% reduction in cost per token between ChatGPT-4 and version 5.4 over just two years, demonstrating massive deflationary pressure in AI compute economics. Despite doubling prices on the newest 5.5 model, customers still receive 20-30% cost reduction per token, suggesting OpenAI's margins are expanding rapidly while maintaining competitive pressure on rivals.

About this episode

In this episode, hosts interview Sarah Friar, CFO of OpenAI, following the company's record-breaking $122 billion fundraising round in March 2025—the largest private or public raise in history, surpassing even Saudi Aramco's $30 billion IPO. Friar revealed OpenAI now commands at least 11% of the global search market, though actual penetration is likely far higher due to measurement methodology favoring traditional search engines. She disclosed the company faces acute compute shortages extending through 2026 and beyond, with supply constraints already emerging for 2030-2032, forcing OpenAI to secure gigawatt-scale data centers years in advance. Friar confirmed OpenAI is developing consumer hardware with designer Johnny Ive, set for unveiling by year-end 2025 and sales in early 2026, described as a paradigm-shifting device that feels "lovable." The conversation explored OpenAI's diversification strategy across multiple cloud service providers, chip manufacturers, and revenue streams now split 50-50 between consumer and enterprise. Friar revealed ChatGPT has reached 900 million weekly users, with paid tiers showing usage rates up to 11x higher than free users. She addressed competitive pressure from Anthropic, which confidentially filed for IPO during the interview, and detailed OpenAI's economics: API business generates revenue per token at an order of magnitude higher than consumer products, yet the company prioritizes consumer distribution to establish ChatGPT as infrastructure. The episode concluded with discussion of OpenAI's advertising strategy, positioning ChatGPT as combining Google's search intent with Meta's demographic targeting, enhanced by user memory and context.

Key takeaways

More stories More from All-In Podcast