OpenAI cuts prices on smaller models as businesses watch costs

OpenAI cut prices on some of its smaller AI models on July 30, putting cost discipline ahead of raw model size as businesses press for measurable returns on AI spending. The lower prices matter most for the models that handle summarization, classification, customer support and routine coding, the kinds of jobs companies want to automate without turning AI into a runaway software expense.

The timing points to a market that has changed quickly. After years of experimentation, many enterprises now want proof that AI tools save time or money, not just impressive benchmarks. OpenAI’s move also fits a broader scramble for customers who are increasingly wary of AI costs, including U.S. companies facing cheaper Chinese rivals and pressure from competitors such as Anthropic. In that environment, price has become a product feature.

Smaller models are often the easiest place to win adoption because they are good enough for high-volume work and cheap enough to roll out across departments. Large enterprises, which have been cautious about moving beyond pilot projects, stand to benefit if lower prices make it easier to deploy AI in customer service, internal search and software development. Developers using OpenAI’s API also get a clearer cost structure, which can matter as much as model quality when applications scale to millions of requests.

OpenAI has been building toward this position for more than a year. On July 18, 2024, the company introduced GPT-4o mini as its “most cost-efficient small model,” saying it would make intelligence “much more affordable.” OpenAI said the model scored 82% on MMLU, outperformed GPT-4 on chat preferences in the LMSYS leaderboard, and came with a 128,000-token context window and an October 2023 knowledge cutoff. OpenAI’s API pricing page lists GPT-4o mini at $0.15 per million input tokens and $0.60 per million output tokens.

The latest price cuts suggest the competition is moving from model capability to commoditization. If smaller models become cheaper across the board, the payoff will go first to customers who buy at scale: enterprises, developers and startups building everyday AI workflows. For vendors, that raises the bar on margins and forces a harder case for why one model is worth more than another.