OpenAI CEO Sam Altman said earlier this week on the Invest Like the Best podcast that OpenAI wants to offer the strongest balance of intelligence, speed and affordability as Chinese start-up Moonshot AI’s Kimi K3 gains attention for powerful AI capabilities. Altman said OpenAI’s goal is to provide the best option for intelligence and price “at every point along the Pareto-optimal frontier,” meaning a trade-off curve where improving one factor usually worsens another.

Altman said OpenAI aims to compete across the full range of AI offerings, from frontier models to smaller and cheaper systems built for faster, more affordable use. He argued that OpenAI’s models currently offer a better deal than Kimi at some levels of latency and performance, saying users get “a better deal today” with OpenAI’s models than Kimi at a particular latency.

Altman said OpenAI makes smaller, cheaper models through distillation, a process where a smaller model learns from a larger one. He also said open-source models will remain important because some users want access to model weights and the ability to modify systems for specific needs. Asked about U.S. regulation and AI competition, Altman said he has “always assumed” that great, cheap models would exist and said OpenAI must become “the greatest and the cheapest.”

Moonshot AI launched Kimi K3 earlier this month as a 2.8 trillion-parameter open-source model. The company said Kimi K3 outperforms leading systems from OpenAI and Anthropic in certain areas and called it the largest open-source model to date, ahead of DeepSeek’s 1.6-trillion-parameter V4 Pro and Zhipu AI’s 744-billion-parameter GLM 5 series. The article noted that parameters indicate model complexity but do not necessarily guarantee better performance. Nvidia has begun shipping H200 AI chips to China after U.S. approval of limited exports.

OpenAI CEO Sam Altman said earlier this week on the Invest Like the Best podcast that OpenAI wants to offer the strongest balance of intelligence, speed and affordability as Chinese start-up Moonshot AI’s Kimi K3 gains attention for powerful AI capabilities. Altman said OpenAI’s goal is to provide the best option for intelligence and price “at every point along the Pareto-optimal frontier,” meaning a trade-off curve where improving one factor usually worsens another.

Altman said OpenAI aims to compete across the full range of AI offerings, from frontier models to smaller and cheaper systems built for faster, more affordable use. He argued that OpenAI’s models currently offer a better deal than Kimi at some levels of latency and performance, saying users get “a better deal today” with OpenAI’s models than Kimi at a particular latency.

Altman said OpenAI makes smaller, cheaper models through distillation, a process where a smaller model learns from a larger one. He also said open-source models will remain important because some users want access to model weights and the ability to modify systems for specific needs. Asked about U.S. regulation and AI competition, Altman said he has “always assumed” that great, cheap models would exist and said OpenAI must become “the greatest and the cheapest.”

Moonshot AI launched Kimi K3 earlier this month as a 2.8 trillion-parameter open-source model. The company said Kimi K3 outperforms leading systems from OpenAI and Anthropic in certain areas and called it the largest open-source model to date, ahead of DeepSeek’s 1.6-trillion-parameter V4 Pro and Zhipu AI’s 744-billion-parameter GLM 5 series. The article noted that parameters indicate model complexity but do not necessarily guarantee better performance. Nvidia has begun shipping H200 AI chips to China after U.S. approval of limited exports.