Archive — US — techcrunch
OpenAI’s o3 model might be costlier to run than originally estimated
OpenAI's new o3 AI model may be more expensive to operate than initially estimated, with costs revised from $3,000 to potentially $30,000 per task. This reflects the increasing computational demands of advanced AI. OpenAI has not yet released pricing for o3, but it may surpass current models like o1-pro, which is considered a close comparison. Efficiency concerns arise, as o3 high required extensive attempts for optimal performance.
Key points
- OpenAI unveiled the o3 reasoning model in December 2023 with partnerships aimed at showcasing its capabilities.
- The Arc Prize Foundation revised the estimated cost to solve an ARC-AGI problem for o3 from $3,000 to $30,000, indicating high operational costs.
- Costs are significant due to o3 high’s extensive computing resource usage, reportedly 172 times that of the low configuration.
- OpenAI has not priced o3 or released it yet, creating uncertainty in operational costs.
- Rumors suggest OpenAI is considering charging enterprise clients up to $20,000 per month for specialized AI agents.
- Efficiency concerns arise as o3 high required 1,024 attempts to achieve its best score, raising questions about its practicality.
Why it matters
- The increasing costs of advanced AI models reflect the trend towards more powerful and resource-intensive technologies in the IT sector.
- Historically, as AI models improve, their operational costs have risen, seen in previous models like GPT-3 and o1-pro.
- The trend highlights the balance between the cost of AI and its efficiency versus traditional human labor in tech.
- In 2025, high operational costs of AI may influence IT budgets and adoption rates among enterprises.
The rising costs associated with OpenAI's o3 model illustrate the challenges of deploying sophisticated AI technologies, raising critical questions about efficiency versus expense as companies weigh their options for AI integration.