OpenAI’s highly anticipated o3 AI model may come with a significantly higher price tag than initially estimated, according to new calculations from the Arc Prize Foundation. The organization, which administers the ARC-AGI benchmark for testing advanced AI reasoning, recently revised its estimates, suggesting that the cost of running o3 could be ten times higher than previously thought.

o3 Model Costs Revised Upward

When OpenAI introduced o3 in December 2024, it was presented as a highly capable “reasoning” AI model with impressive problem-solving abilities. However, new findings indicate that the computing costs to operate o3 might be far steeper than expected.

• Initially, the Arc Prize Foundation estimated that the best-performing version of o3, known as o3 high, cost around $3,000 to solve a single ARC-AGI problem.
• The latest calculations now suggest that the actual cost could be closer to $30,000 per task—a tenfold increase from the original estimate.

This major revision highlights how expensive cutting-edge AI models may be to run, especially in their early stages.

How Does This Compare to Other OpenAI Models?

OpenAI has not yet publicly disclosed the official pricing for o3, nor has it released the model for commercial use. However, the Arc Prize Foundation suggests that OpenAI’s o1-pro model pricing could serve as a useful comparison.

• o1-pro is currently OpenAI’s most expensive model.
• The Foundation believes o3 will require even more computational power, making it one of the costliest AI models OpenAI has developed.
• Mike Knoop, co-founder of The Arc Prize Foundation, noted that o1-pro’s pricing structure might be a more accurate benchmark for o3 costs due to the amount of test-time compute involved.

“We believe o1-pro is a closer comparison of true o3 cost due to the amount of test-time compute used,” Knoop told TechCrunch.

Computational Demands of o3 High

One of the biggest contributors to o3 high’s potential cost is its enormous computing requirements.

• According to the Arc Prize Foundation, o3 high required 172 times more compute power than o3 low, the least resource-intensive version of the model.
• Additionally, o3 high needed 1,024 attempts per task in ARC-AGI to achieve its best score.

These factors suggest that o3’s computational efficiency may be lower than expected, which could drive up operational costs for businesses and researchers hoping to use it.

Is OpenAI Planning Expensive AI Pricing for Enterprises?

There have already been reports indicating that OpenAI may be planning premium-priced AI solutions for enterprise clients.

• In early March, The Information reported that OpenAI could charge up to $20,000 per month for specialized AI agents, such as software developer agents designed for enterprise-level automation.
• Given the computing intensity of o3 high, it’s possible that similar high-cost plans may be in store for its deployment.

While some argue that even OpenAI’s priciest AI models would still be cheaper than hiring human experts, critics point out that the efficiency of these models is still uncertain. AI researcher Toby Ord highlighted in a recent post on X (formerly Twitter) that o3 high’s high number of attempts per task could make it less efficient than expected.

What This Means for OpenAI’s AI Strategy

If the revised $30,000 per task estimate holds true, o3’s long-term viability could depend on major efficiency improvements.

• OpenAI may need to optimize o3’s compute usage to make it more cost-effective for real-world applications.
• Businesses considering OpenAI’s AI solutions will need to weigh performance benefits against operating expenses.
• The competition among AI leaders—including Google DeepMind, Anthropic, and xAI—could push OpenAI to adjust its pricing strategy to remain competitive.

For now, OpenAI has not officially responded to the revised cost estimates, but as the company moves closer to releasing o3, its final pricing will be closely watched by the AI industry.

Get the Latest AI News on AI Content Minds Blog