OpenAI will publicly release its GPT-5.6 Sol, Terra and Luna models on Thursday, the strongest it has built.
The company says the family is markedly better at coding, biology and cybersecurity, the areas where enterprise buyers spend most.
The launch raises an obvious question: can OpenAI use this generation to pull back ahead of Anthropic, the rival that overtook it this year?
The answer depends on which race is being run, and on how much raw computing power decides the outcome.
How the lead changed hands
For most of the AI era, OpenAI was the clear front-runner.
That changed in early 2026, when Anthropic's annualised revenue reached about $30 billion, moving past OpenAI's roughly $25 billion.
The shift was sharpest in the enterprise market, where Anthropic captured the majority of first-time corporate AI spending after trailing OpenAI only months earlier.
Its coding tool, Claude Code, drove much of the surge, and about 80% of its revenue now comes from business customers, against roughly 40% at OpenAI.
Crucially, that reversal was not won with superior compute.
It was won on focus, efficiency and stickier enterprise contracts, a reminder that the recent battleground rewarded discipline rather than the biggest training run.
Why compute still matters
Yet compute is exactly where OpenAI holds its clearest advantage.
The company has committed to about $600 billion of compute spending through 2030, revised down from an earlier $1.4 trillion, dwarfing anything a comparable business has attempted.
Industry projections put its annual training bill near $121 billion by 2030, against roughly $30 billion for Anthropic.
That war chest is funded by a $122 billion round in March that valued OpenAI around $852 billion, with Amazon, Nvidia and Microsoft among the backers.
More compute buys larger models, faster iteration and the capacity to serve hundreds of millions of consumers, and GPT-5.6 is the first big test of what that spending now delivers.
If the new models open a clear capability gap in coding and other enterprise tasks, compute will have done its job.
Where the strategy strains
The difficulty is that spending on this scale carries its own risk.
OpenAI is projected to lose about $14 billion in 2026 and is not expected to turn cash-flow positive until around 2030.
Its economics still lean on a vast free consumer base that generates heavy inference costs without matching revenue.
Anthropic, by contrast, is targeting software-like margins and a path to positive cash flow as soon as 2027, backed by its own compute deals with Amazon, Google and others.
So the compute advantage is real, but it is not free, and it does not automatically convert into the enterprise revenue where the lead actually moved.
Which race is being run
This is why the question splits in two.
On frontier capability and consumer reach, OpenAI's compute budget gives it a genuine chance to reassert dominance, and a strong GPT-5.6 could do exactly that.
On the enterprise and revenue metrics where Anthropic pulled ahead, compute helps but does not settle the contest, because those buyers reward reliability, integration and cost as much as headline performance.
The government dimension adds another layer, with both companies now routing frontier releases through a White House framework that asks developers to share cutting-edge models before wider launch.
That process, which briefly restricted access to Anthropic's Claude Fable 5 and Mythos 5 models, could shape how quickly either firm can turn a technical edge into a market one.
The likely outcome
The most probable result is not a single winner but a divergence.
OpenAI can plausibly lead on model capability and consumer scale while Anthropic holds its grip on enterprise revenue.
Compute may well supercharge OpenAI's surge, but the past year suggests it is a necessary condition for pulling ahead, not a sufficient one.
