Updated
Updated · South China Morning Post · Jul 20
Moonshot’s 2.8 Trillion-Parameter Kimi K3 Narrows US AI Gap to Weeks
Updated
Updated · South China Morning Post · Jul 20

Moonshot’s 2.8 Trillion-Parameter Kimi K3 Narrows US AI Gap to Weeks

3 articles · Updated · South China Morning Post · Jul 20

Summary

  • Moonshot’s Kimi K3 has cut the lead of top US AI models from an estimated six to eight months to roughly weeks, according to American Enterprise Institute analyst Ryan Fedasiuk.
  • The Beijing startup’s open-weight model combines 2.8 trillion parameters with native vision and targets long-horizon coding, knowledge work and complex reasoning tasks.
  • Benchmark results intensified Silicon Valley concern: Kimi K3 outperformed GPT-5.6 Sol and Claude Fable 5 in some tests, including Arena AI’s front-end coding ranking.
  • Pricing adds to the pressure, with Kimi K3 still significantly cheaper than US rivals even though it costs more than other Chinese models.
  • The release reinforces a broader view that Chinese AI developers are closing on US frontier systems faster than many in the industry had assumed.

Insights

As China's Kimi K3 challenges top US models, is the high-cost, closed-source AI business model now obsolete?
Firms flee to open-source AI for data privacy, but can we trust models allegedly trained on stolen data?
With AI valuations hitting bubble territory, can any company justify a $31.5B price tag in a commoditizing market?