An open model that thinks, and what that combination means
Moonshot released a downloadable reasoning model yesterday. Reasoning used to be the thing the closed models kept for themselves.
On 6 November 2025 Moonshot AI released Kimi K2 Thinking, an open-weight model built for step-by-step reasoning. It is the clearest sign yet that the last remaining gap between open and closed models is closing.
Why reasoning was the last thing to arrive
When reasoning models appeared — this blog covered the first of them in autumn 2024 — they were expensive to build and expensive to run, and they were the argument for paying for a frontier provider.
Ten months later that argument is weaker. The capability that justified the premium is now available in a form you can download, and the pattern from the 10 September post holds: every distinction between the two camps has eventually dissolved.
What this changes for a Swiss SME, in plain terms
Directly, very little. You are no more likely to run a large reasoning model yourself this quarter than you were last month, and the 4 September assessment of what self-hosting actually involves is unchanged.
Indirectly, one thing: it strengthens your position in a conversation with a supplier. When capability is available from several sources including free ones, a supplier’s price has to be justified by service, contract and support rather than by exclusive access to something clever.
That is a better basis for a supplier relationship than the one that existed a year ago.
The thing worth resisting
The temptation to re-open a decision you already made. If you chose a provider in spring on the basis of data location, contract and support, none of those reasons have been weakened by this release.
Three more launches are expected before the end of the month. The 21 November post will be about how not to react to all of them, which is a subject the calendar is about to make urgent.