Moonshot AI launched the Kimi K3 API on July 16, promising to publish the model's weights on the 27th of the same month. According to reporting from the period, the Chinese model wiped $314 billion from the pre-IPO valuations of two American frontier companies.
The valuation figure should be read carefully, since private company valuation is an estimate. The verifiable fact is different: a competitive model with published weights changes the arithmetic for anyone selling access by subscription, because it establishes a quality floor available without a monthly fee.
The eleven-day gap between the paid API and the weight release is a strategic choice. It gives the press time to measure the model through the commercial route before it becomes a public good, which usually yields a more favourable comparison.
For anyone operating systems, a model like this has an immediate practical effect: it becomes a fallback you can run on your own hardware, without depending on approval, quota or someone else's price.
The contrast with the week is stark. While American companies discussed voluntary release standards with their government, the competition published the weights and made the discussion inapplicable to its own product.