On the first night of Kimi K3 open source, at least 18 American companies quickly deployed Kimi K3. After the weight of Kimi K3 was officially launched on Hugging Face open source last night, a very ironic drama immediately unfolded - although the U.S. government and Anthropic and other companies are spending a lot of money to lobby to block China's large open source model, at least 18

On the first night of Kimi K3 open source, at least 18 American companies quickly deployed Kimi K3. After the weight of Kimi K3 was officially launched on Hugging Face last night, a very ironic drama immediately unfolded - although the U.S. government and Anthropic and other companies are spending a lot of money to lobby to block China's open source model, at least 18 American companies completed the deployment as soon as possible and provided commercial services to the outside world. This list compiled by netizen Ding exposes the rift between Washington’s political narrative and Silicon Valley’s pragmatic reality. Top performance coupled with ultimate cost-effectiveness, commercial rationality overwhelms political statements. The actual logic is actually not complicated: K3 is currently recognized as one of the three most powerful models in the world, while the OpenAI and Anthropic models are not only closed source, but also expensive and have many restrictions. Open source deployment allows enterprises to get rid of their dependence on closed source APIs, allowing them to customize and optimize themselves, significantly reducing usage costs. Of course, deploying K3 is far from "download and run." Although this behemoth with 2.8 trillion parameters can run with at least 8 AMD or NVIDIA AI graphics cards, Dark Side of the Moon officially recommends a 64-card cluster solution. It is estimated that the cost to truly provide smooth API services will start at about 10 million yuan, and to achieve a commercial-level smooth experience will require an investment of about 30 million yuan. In other words, among the 18 American companies that quickly took action, each one calculated their accounts carefully - instead of endlessly paying closed-source manufacturers based on tokens, it is better to spend one time on infrastructure costs and truly hold the right to use the model in their own hands. In the second half of the commercialization of large models, controlling inference costs is becoming a more important strategic decision than signing exclusive cooperation. via AI News (author: AI Base)