China's Moonshot released Kimi K3.1 Pro, a version of Kimi K3 with much more post-training, now available through its apps and API. It costs 2 dollars per million input tokens and 10 dollars per million output tokens, about half the price of the most expensive closed models. It scores 55 on the Artificial Analysis index, just behind the leading model, with large gains in coding and tool-use tasks. Moonshot says it will publish the weights under an open licence on 4 November, while reports of an influencer campaign without paid-partnership labels have raised questions.

What it means: For Greek software companies, Kimi K3.1 Pro is another sign that the gap between open and closed models is closing fast. If the weights really arrive under an open licence, there will be one more option to run a top-tier model on your own servers or on a European cloud, which matters for anyone handling sensitive customer data and wanting control over where it is processed. Until then, access goes through Moonshot's own service, which according to its policy uses servers outside the EU, so a careful GDPR check is needed before any personal data is sent. The model is very large, so running it locally will require serious hardware and is not a fit for every small team. The practical advice is simple: test it on your own tasks, measure the cost per completed piece of work rather than per token, and do not rely on benchmark scores or enthusiastic posts. The influencer allegations are a reminder that reviews of AI tools on social media are not always independent. The licence text has not been published yet either, so the terms for commercial use and for public-sector clients still need to be clarified.

Read the original on VentureBeat

Related service: Custom software

Related stories