OpenAI’s o3 Model Costs Up To $3,500 Per Query, Numerama Reports

OpenAI’s o3 model, announced on December 20, 2024 as part of OpenAI’s 12 days of announcements, is described as a frontier model that excels at mathematical tests and could one day solve puzzles too complex for humans by thinking for several minutes and cross-referencing billions of data points.

Numerama reports that a query to the best version of o3 easily exceeds $1,500, with the cost described as $3,500 depending on methodology, and says the most powerful version of o3 may never come out.

Challenges

Arc Prize, which had access to early versions to run benchmarks, says a query to o1-mini currently costs about twenty cents, while o1 in its most expensive version costs about $5 per question.

Numerama adds that the first version of o3 tested by Arc Prize, costing $17–$20, has six computation iterations and can answer 75.7% of the most complex questions, while GPT-4o was limited to 5%.

The boosted version, at $1,600, uses 1,024 iterations and OpenAI is said to break its own record with 87.5% correct answers, raising the question of whether o3 will be reserved for researchers rather than the general public.

Developpez reports that corporate spending on AI is skyrocketing, with some organizations paying more for computing power than for human labor, as executives question whether the AI boom is sustainable.

The article ties the shift to AI business models that reorient monetization around the token as the central billing unit, with usage limitations, segmented features, and higher prices redefining relationships among providers, businesses, and developers.

Democracy Now!

It describes Oracle’s restructuring as a glimpse of how investments in AI are redefining corporate priorities, anchored by a cloud computing deal with OpenAI valued at around $300 billion to build and expand data centers for advanced AI workloads.

Developpez says Oracle’s total borrowings are estimated to have surpassed $100 billion, and that sizable capital expenditures pushed its free cash flow into negative territory, while layoffs are framed as rebalancing with analysts estimating savings of between $8 billion and $10 billion.

The piece also quotes Bryan Catanzaro, vice president of applied deep learning at Nvidia, saying his team’s computing costs now far exceed personnel expenses, and notes that at Uber the chief technology officer is said to have exhausted the company’s 2026 AI budget ahead of schedule due largely to token-related expenses.

Democracy Now! revisits an interview with longtime technology reporter Karen Hao, author of Empire of AI, describing her account of the political and economic power of artificial intelligence companies, especially Sam Altman’s OpenAI.

“In our July Fourth special broadcast, we revisit our interview with longtime technology reporter Karen Hao, author of Empire of AI, which unveils the accruing political and economic power of artificial intelligence companies — especially Sam Altman’s OpenAI”

Democracy Now!

Hao compares the actions of the AI industry to colonial powers and says, “The empires of AI are not engaged in the same overt violence and brutality that marked this history,” while adding that they “seize and extract precious resources” including “the work of artists and writers” and “the land, energy, and water required to house and run massive data centers and supercomputers.”

In the same broadcast, Hao says Silicon Valley’s scale-at-all-costs approach involved “blowing up the amount of data and the size of the computers” needed for training, with “the full English-language internet” fed into models and supercomputers running “tens of thousands, even hundreds of thousands” of chips.

She provides two stats on energy and water, citing a McKinsey report that in the next five years, based on the current pace of AI computational infrastructure expansion, the world would need to put as much energy on the global grid as “two to six times the energy consumed annually by the state of California.”

For freshwater, Hao says data centers need to be trained on freshwater and that Bloomberg analysis found “two-thirds of them are being placed in water-scarce areas,” placing them in communities that do not have access to freshwater.