<meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <title>HF Demo</title> <style> body { margin: 0; font-family: Arial, sans-serif; } header { background-color: #2c3e50; color: white; text-align: center; padding: 20px; font-size: 28px; font-weight: bold; } </style>

OpenAI’s Brockman Says Kimi K3 Is ‘Pretty Good,’ Won’t Confirm Distillation

OpenAI's Brockman Says Kimi K3 Is 'Pretty Good,' Won't Confirm Distillation

In a Bloomberg interview on Tuesday, OpenAI president Greg Brockman acknowledged the quality of Chinese rival Moonshot AI’s Kimi K3, describing it as "pretty good." He added that it is too early to determine if Moonshot distilled from OpenAI's outputs, a practice previously accused by OpenAI.

Hours earlier, OSTP Director Michael Kratsios had leveled specific accusations against Moonshot, alleging access to banned Nvidia chips and large-scale distillation against American AI models. These claims came as part of growing concerns expressed by the US government about potential intellectual property theft from Chinese AI startups.

Nvidia CEO Jensen Huang pushed back, describing Chinese open-source models as "excellent" and urging the government to allow American firms to use them.

The distillation charge has been building for months. In 2026, Greg Brockman, OpenAI's president, maintains a nuanced stance on Moonshot AI's Kimi K3. While acknowledging its 'pretty good' performance, he avoids confirming distillation practices, emphasizing that it's too early to judge. This cautious approach reflects the evolving landscape of AI, where accusations of intellectual theft are met with a focus on infrastructure investments and the US's compute power advantage. Brockman highlights the four-month gap between China's model development and the US, despite Kimi K3's impressive specifications. The commoditization of models, driven by Moonshot's pricing strategy, further complicates the debate, as investors eye the model layer's potential. Read the full article here In February, Anthropic accused Moonshot, DeepSeek, and MiniMax of stealing knowledge from its Claude models by running millions of exchanges through fraudulent accounts. OpenAI has made similar accusations, and now the White House appears to be treating these allegations as a matter of national security.

Brockman framed OpenAI’s position around its infrastructure investments, arguing that the US has a “huge advantage” due to billions invested in compute power. He estimated China lags behind the US by roughly four months in overall model development despite releasing Kimi K3, a nearly three-trillion-parameter open-weight model, which has matched or beat the best American systems on several benchmarks.

Moonshot is pricing K3 at a fraction of what US labs charge, accelerating the commoditisation of the model layer that investors have been eager to see.

Leave a Reply