The competition in artificial intelligence has shifted. It is no longer only about who builds the biggest model. It is about who can sell a model cheap enough and fast enough to run inside a product.
OpenAI’s latest move lands on that second question. On October 8, the company released an “Ultrafast” version of GPT-6.1 Sol, its latest model, across its API, Codex and ChatGPT Work offerings. The version is priced at $12 per million input tokens and $60 per million output tokens — a discount to the full model that buys speed over depth.
The pricing is the story. Input tokens are what a model reads; output tokens are what it writes. At $12 in and $60 out, the Ultrafast tier undercuts the flagship and signals that OpenAI expects a large share of its business to come from tasks where a fast, adequate answer beats a slow, brilliant one.
That is the economics of agents. Systems that browse, call tools and run workflows make many small model calls in sequence. Each call must return quickly, and every token of output costs money. A cheaper, faster model changes the arithmetic of whether an agent is worth running at all.
OpenAI is selling into a market that has learned to comparison-shop. Rivals led by Anthropic and Google push their own models at competing price points, and developers routinely route tasks to whichever model clears the quality bar at the lowest cost. A new tier is a bid to keep those developers on OpenAI’s platform.
The Ultrafast label also points to a technical trade-off. Faster inference generally means the model does fewer steps of reasoning before answering. For summarization, drafting, classification and routine coding, that is often enough. For hard research or long planning, a slower model with more reasoning time still wins. OpenAI is betting that a large share of real-world use is the first kind.
The rollout spans three channels, which says something about where OpenAI thinks the demand is. The API reaches developers. Codex reaches programmers inside their editors. ChatGPT Work reaches businesses, where the company has been pushing to convert its consumer success into enterprise contracts.
That enterprise push is central to OpenAI’s revenue ambitions. The company has told investors it expects annualized revenue to climb steeply, and business subscriptions are a steadier source than consumer chat. A cheaper tier lowers the cost for companies that want to wire the model into their own operations at scale.
The “Sol” name signals a product family rather than a single release. OpenAI has split its flagship line into variants tuned for different jobs, the way chipmakers sell a family of processors at different clock speeds and price points. The Ultrafast tier is the budget option in that lineup, aimed at volume over peak capability.
The pricing also reflects a maturing market. In the early years, models competed on benchmark scores. Now they compete on tokens-per-dollar, latency and reliability, the same metrics that govern any infrastructure business. The companies that win will be the ones that can cut costs fastest while holding quality.
OpenAI has a reason to move quickly. Its position at the top of the market is no longer taken for granted, and price cuts from competitors have repeatedly forced the field to follow. Releasing a fast, cheap tier is both an offense and a defense: it captures low-latency demand and pressures rivals to match.
For customers, the choice is now a menu rather than a single model. A developer can pick the full GPT-6.1 Sol for hard problems and drop to Ultrafast for everything else, mixing tiers by task. That flexibility is the direction the whole industry is heading.
The question is whether speed and price, without a flagship’s full reasoning, will hold the customers OpenAI is courting. The Ultrafast version is cheaper for a reason, and the reason will show up in the hardest tasks. For the routine work that fills most enterprise pipelines, though, it may be enough.
The timing is deliberate. Model releases now arrive in rapid succession, and each new tier is an attempt to hold developers who might otherwise drift to a cheaper rival. OpenAI’s pricing moves are watched as closely as its model names, because price is the fastest way to shift share in a market where switching costs are measured in a few lines of code.
That, in the end, is the bet. Most of what a model does inside a company is not frontier research. It is triage, drafting, classification and glue code — work that needs to be quick and inexpensive more than it needs to be deep. OpenAI has now put a price on that distinction, and the market will say whether it got the number right.
