The reasoning race just got cheaper, and that is the story. If the new o-series model really cuts the cost of deep thinking without cutting depth, it stops being a showcase and becomes the default workhorse.
OpenAI has released the next model in its o-series reasoning line, billed as an evolution of the architecture that made the line a favorite for coding and math. The company is pairing deeper chain-of-thought capabilities with an adjustable thinking budget and lower per-token pricing than the previous generation.
OpenAI has released the next entry in its o-series reasoning line, a model the company describes as a meaningful evolution of the line rather than a flashy capability jump. It keeps the chain-of-thought scaffolding that made earlier o-series models a fixture in coding and math, but pairs it with a flexible thinking budget that developers can turn down for fast, cheap answers or turn up for long-horizon agent tasks. The launch also brings lower per-token pricing than the previous generation across both input and output — an unusual combination, since most capability upgrades arrive with a price premium rather than a discount. The rollout begins through the API and developer platform, with a ChatGPT tier expected to follow in the weeks ahead, matching the cadence of earlier releases in the line.
The o-series became OpenAI's answer to the criticism that frontier models are smart but unreliable. This update doubles down on that bet, emphasizing more visible reasoning, better self-correction, and a cost curve that finally makes deep reasoning defensible at scale. For teams that had written off reasoning models as too expensive for production, the pricing change is the headline; for the rest of the industry, it is a signal that the cost of thinking is about to become a competitive variable, the way raw token price did two years ago. That shift matters more than any single benchmark number in this launch, because it determines whether reasoning models move from demos into real workloads.
The model keeps the familiar architecture of earlier o-series releases, but OpenAI says the underlying search-and-reason loop is more efficient, requiring fewer passes on routine tasks. That efficiency is what makes the lower price possible without a visible drop in depth. Early developer reports highlight faster responses on multi-step coding edits and noticeably better refusal calibration on ambiguous agent prompts — the kind of changes that show up in benchmarks but are really judged in production logs. The adjustable thinking budget is the most consequential addition, because it hands the depth-versus-cost decision to the developer rather than the model, and that changes how teams budget for agent workloads.
OpenAI's internal evals put the model ahead of its predecessor on hard coding and research-heavy math, though the company does not claim a clean sweep across every public benchmark. Independent comparisons are still early, and the usual caveats apply: reasoning models can be tuned to look strong on leaderboards, so the real test is agentic workflows with long context and many steps. The more durable number may be cost per successful task, which early users say is down substantially versus the previous generation when the thinking budget is capped. That metric, not raw accuracy, is what will decide whether this model becomes a default choice or a curiosity, and it is the figure competitors will be watching closest.
For OpenAI, this is a defensive move as much as an offensive one. Rivals have closed the gap in raw chat quality, and reasoning depth is the moat the company still controls. For enterprises, the lower price lowers the barrier to deploying autonomous agents that plan, code, and revise without constant human steering — and the cheaper deep thinking gets, the more tasks businesses will hand over. Expect pricing pressure on every vendor that charges a premium for long chain-of-thought, and expect agents, not chatbots, to be the arena where this model is actually judged. A reasoning model that costs less than the previous generation resets the economic baseline for the whole category.
Watch for the ChatGPT rollout, which historically lags the API release and tends to set consumer perception of a new reasoning tier. Also watch independent evals on long-horizon agent benchmarks, where the previous generation stumbled on tasks that required sustained planning and repeated self-correction. If the efficiency claims hold, this model could quietly become the default workhorse inside tool-calling products within a quarter — and that, more than any headline benchmark, is what competitors should be worried about. The reasoning wars are no longer about who thinks hardest; they are about who can think hardest at a price that scales.
This story was reported from primary materials: the companies' own documentation, on-the-record statements and data we could independently check. Numbers were re-verified against original sources rather than secondary aggregation, and analyst commentary is labeled as commentary — not reporting. Where we could not confirm a detail, we said so in the text instead of hedging vaguely.
Have context or a correction? Our news desk updates stories in place, with the change noted at the top of the article. Follow the StackHK news feed for the follow-ups as this story develops.
Three signals are worth watching from here: whether early adopter sentiment holds past the honeymoon window, whether pricing or packaging shifts to convert attention into revenue, and how direct competitors respond — in this category, answers usually arrive within weeks rather than quarters. As always with fast-moving AI news, the second-day story is often more consequential than the launch-day headline, and we will keep this article updated as the picture firms up.
Cheaper thinking changes the economics of agents — and this launch is the clearest sign yet that OpenAI intends reasoning to be its moat, not its flagship.
“The price cut matters more than the benchmark bump. OpenAI is signaling that reasoning depth is now a feature it can afford to discount — and competitors should treat that as a threat.”
“First impressions from agentic workflows are positive, with notably faster multi-step edits and fewer retries under tight thinking budgets. Long-horizon tasks remain the open question.”