OpenAI’s o3 AI Model: High Costs and Mixed Results

When OpenAI launched its o3 “reasoning” AI model in December, the anticipation was palpable. The partnership with the creators of ARC-AGI—a benchmark designed to evaluate highly capable AI—promised to showcase o3’s groundbreaking capabilities. Fast forward a few months, and the results have been revised, revealing a less-than-stellar picture than initially expected.

Shocking Cost Revisions

Just last week, the Arc Prize Foundation, which oversees the ARC-AGI benchmark, updated its previous estimates for using o3. Initially pegged at around $3,000 per task for the highest-performing configuration, dubbed o3 high, the new estimate has skyrocketed to possibly $30,000. That’s a tenfold increase that raises eyebrows about the true cost of cutting-edge AI technology.

This revision is significant. It underscores the exorbitant expenses associated with some of today’s most advanced AI models, at least during their early stages. OpenAI has yet to set a pricing structure for o3 or even officially launch it. However, the Arc Prize Foundation has suggested that the costs associated with OpenAI’s o1-pro model might be a reliable indication of o3’s potential pricing.

Pricing Comparisons: o1-pro vs. o3

To put it into perspective, o1-pro is OpenAI’s costliest model to date. "We believe o1-pro is a closer comparison of true o3 cost due to the amount of test-time compute used," Mike Knoop, co-founder of the Arc Prize Foundation, told TechCrunch. He emphasized that o3 is still labeled as a “preview” on their leaderboard to reflect the uncertainty surrounding its pricing until formal announcements are made.

Given the computing demands of o3, it’s likely that the high price tag wouldn’t be surprising. The Arc Prize Foundation reports that o3 high consumed 172 times more resources than o3 low, the most efficient version, when addressing ARC-AGI tasks.

The Potential for High-End Pricing

Rumors have swirled for a while about OpenAI’s plans for pricey options geared towards enterprise clients. Reports from early March hinted at possible charges of up to $20,000 per month for specialized AI “agents”—such as those designed to act like a software developer.

While some may argue that even OpenAI’s most expensive models could be less costly than hiring a full-time staffer or contractor, AI researcher Toby Ord raised an important point on social media. He noted that this efficiency comes into question, as o3 high required 1,024 attempts for each ARC-AGI task to obtain its best score. This raises questions about how economically efficient these AI systems truly are when stacked up against human talent.

In Conclusion

The journey of OpenAI’s o3 model is certainly one to watch, given its staggering costs and questionable efficiency. As technology in the AI space continues to evolve rapidly, we can only imagine how these factors will influence its deployment in various sectors.

The AI Buzz Hub team is excited to see where these breakthroughs take us. Want to stay in the loop on all things AI? Subscribe to our newsletter or share this article with your fellow enthusiasts.