China’s Kimi K3 has climbed to first place on the Frontend Code Arena, prompting former White House crypto czar David Sacks to warn that US regulation could weaken America’s position in the AI race.
Summary
- Kimi K3 topped the Frontend Code Arena and posted strong results across other AI benchmarks.
- David Sacks warned that strict US rules could weaken America’s position against China.
- Moonshot AI plans to release Kimi K3’s open weights by July 27.
Sacks described the benchmark result as concerning because Kimi K3 also performed close to leading models across several other evaluations. He argued that restrictions on data centers, state-level rules and proposed federal reviews could slow American developers while Chinese companies continue building advanced systems.
“This is how you lose the AI race,” Sacks wrote.
According to Sacks, the United States became a technology leader during the internet era by allowing companies to develop products without seeking government approval in advance. He believes Washington should follow a similar approach to AI while using targeted rules to address specific safety risks.
Kimi K3 strengthens China’s position in advanced AI development
Moonshot AI built Kimi K3 with 2.8 trillion parameters, a one-million-token context window and native multimodal capabilities. The company designed the model for lengthy coding assignments and agent-based workflows that require systems to complete several connected tasks.
Moonshot AI reported that its Kimi Delta Attention system delivers decoding speeds up to 6.3 times faster when processing one-million-token contexts. Its Attention Residuals technology also improves training efficiency by about 25% while adding less than 2% to the total cost, according to the company.
Beyond the Frontend Code Arena, Kimi K3 recorded an Elo score of 1,668 on GDPval v2. The reported result placed it above GLM-5.2, GPT-5.5, and Claude Opus 4.8, although Claude Fable 5 retained a higher score.
On AutomationBench-AA, the Chinese model achieved a 53% score and took first place in the test for agent-led software-as-a-service workflows. Results published through nextjs.org/evals also placed Kimi K3 ahead of Fable while showing comparable task success in less time, according to the report.
Moonshot AI has released Kimi K3 through Kimi.com, Kimi Work, Kimi Code and the Kimi API. The company expects to make the model’s open weights available by July 27, giving developers another way to test and adapt its capabilities.
US restrictions face scrutiny as foreign AI models advance
Sacks linked Kimi K3’s performance to an intensifying debate over how Washington should oversee frontier AI. His criticism covered proposals for federal pre-approval, limits on data center construction and the growing number of AI rules introduced by individual states.
At the federal level, the US government has approved limited access to Anthropic’s Claude Mythos 5 for roughly 100 businesses and agencies. In a letter to Anthropic co-founder Tom Brown, Commerce Secretary Howard Lutnick said selected partners could use the model under specific safeguards.
“I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.”
While Sacks acknowledged that AI safety requires government attention, he argued that other countries would not copy American restrictions. Under his assessment, rules that delay infrastructure construction or model development could leave US companies competing against foreign laboratories operating under fewer limits.
President Donald Trump has also used competition with China to support his technology and digital-asset agenda. Trump has urged the Senate to pass the CLARITY Act, arguing that delayed crypto legislation could allow China to gain ground, while his administration has promoted the United States as the “crypto capital of the world.”
Sacks’ response to Kimi K3 applies the same competitive argument to artificial intelligence: address clear risks without placing approval barriers in front of US developers.
