OpenAI shipped o3 to ChatGPT Plus, Pro, and Team users today, and the number that deserves your attention is not the benchmark score — it is the price. o3 is now available at the same subscription tier as GPT-4o. Six months ago, frontier reasoning required a separate API plan at a price that kept it in the hands of researchers and well-funded startups. Today it is a radio button in the ChatGPT model picker. That is the actual news. Reasoning at scale — the ability to decompose a hard problem, hold its parts in working memory, check its own conclusions, and try again — has just been commoditised for 100 million ChatGPT users.

The benchmark gap that o3 closes is real and was quietly shaping who could do what with AI. GPT-4o and Claude 3.5 Sonnet were exceptional at language, fluency, and instruction following. But problems that required genuine multi-step reasoning — competition-level maths, PhD-tier science questions, complex legal analysis — still stumped them in ways that frustrated power users. o3 was the model that changed this on the research circuit, beating human averages on the ARC-AGI benchmark for the first time. The debate about whether that was a general reasoning breakthrough or an expensive brute-force pass is, for most practical purposes, now irrelevant. The thing works. It is in the product.

The business implications are immediate and will take time to absorb fully. Consulting firms that competed on the ability to synthesise complex information and produce strategic recommendations now share their core competency with a $20/month subscription. Legal document analysis, due diligence, financial modelling, technical architecture review — the categories of knowledge work that required expensive human expertise specifically because they required sustained reasoning — are now reachable by o3 in a first pass. The first pass will not always be right. But it is now a credible first pass, and it is free to run at the margin.

The competitive signal this sends to Anthropic, Google, and the open-source community is unambiguous: reasoning is no longer a differentiated tier. It is table stakes. The race now moves to reasoning that is grounded in up-to-date information, reasoning that can use tools reliably, reasoning that does not hallucinate its way through a calculation. o3's arrival in ChatGPT does not end the model wars. It ends Phase One. Phase Two, which starts today, is about what you do with reasoning once everyone has it.