OpenAI has implemented an 80% price reduction for its o3 model without altering the model itself or compromising its performance. Despite rumors suggesting that the price cut was due to replacing o3 with a smaller, distilled, or quantized version, multiple OpenAI team members and independent assessments by ARC AGI have confirmed that the model remains unchanged. Retesting of the o3-2025-04-16 model on ARC-AGI benchmarks showed no difference in performance compared to previous results. OpenAI attributed the price cut to improvements in inference stack optimization and operating on previously high profit margins, rather than any downgrade in model capability. This development highlights rapid advancements in AI efficiency and cost-effectiveness.
ARC AGI confirms that the o3 is the same model after the price reduction. (Some speculated the price drop means the model was substituted for a lesser "distilled" version) https://t.co/UooNGXOLsF
o3 cost reduction is proof that AI is evolving rapidly. It's becoming more efficient and cost-effective. There's no slowing down. https://t.co/D5hi7SV1Bg
It is now clear that the current o3, which is 80% cheaper, is the same o3 that we have been using the whole time. It remains exciting to see how they were able to reduce the inference costs so much. https://t.co/ic6ETpOLqA