OpenAI releases o4-mini-high: frontier reasoning at 60% lower cost than o3

Think of it like hiring a highly skilled consultant who charges a fraction of what a senior partner at a top firm would cost, but delivers nearly identical results. That is essentially what OpenAI has done with o4-mini-high.

This model belongs to the family of "reasoning models" — AI systems that do not just reply instantly but instead pause to think through a problem step by step before answering. This internal deliberation is called extended thinking, and in o4-mini-high it can span up to 32,000 tokens, allowing the model to work through genuinely complex problems in depth.

What makes this release notable is that the model can use external tools while it is still in the middle of reasoning. It can run code, search for information, or query a database not after it finishes thinking, but as part of the thinking process itself. This mirrors how a skilled engineer might pull up documentation or run a quick test while working through a problem mentally, rather than planning everything in advance and then acting.

Performance on official benchmarks is very close to o3, OpenAI's most capable reasoning model, but the API cost is 60% lower. This means developers and companies can build intelligent applications — coding assistants, document analysis tools, complex question-answering systems — without paying premium prices for every query.

The practical effect is that high-quality AI reasoning becomes accessible to a much wider range of projects and budgets, not just large enterprises with deep pockets.