Moonshot Launches Kimi K3.1 Pro With Hosted Access Now and Weights Promised by November 4

What is available now, and what is still a promise

Moonshot introduced Kimi K3.1 Pro on its blog as the first model in its K3.1 series, describing it as its most capable 3T-class open-weight model to date with a one-million-token context window. The company also introduced Kimi K3.1 Pro Highspeed, which it says delivers two to three times the speed of K3.1 Pro. According to the announcement, both are available today on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API, and the full model weights will be released by November 4, 2026. The public Kimi API platform lists both models, which corroborates the hosted availability claim at the level of a product listing.

The distinction between those two statements is the story. Hosted access is a present, verifiable fact: the models are listed on the vendor's platform today. The weights are a future commitment with a stated deadline, not a released artifact. As of this writing, the weights are not published, and no independent verification of a weight release exists. A model described as open-weight is not open-weight in practice until the files, license, and provenance are actually downloadable and inspectable.

What an independent evaluator reports

Artificial Analysis, a third-party benchmarking organization, published its own assessment of the hosted model on October 23, 2026. It reports that Kimi K3.1 Pro scored 55 on its Intelligence Index, an eleven-point gain over the company's previous Kimi K3, and states that the weights are currently not published but expected soon after launch. The evaluator notes that once the weights are released, it expects the model to become the highest-scoring open-weights model on its index.

These are the evaluator's own measurements on its own test suite, not universal rankings and not results from your workloads. The vendor's blog also cites its own benchmark figures, including a 65.7% score on Terminal-Bench 4.0. Vendor numbers and evaluator numbers come from different setups and should be read as directional signals rather than guarantees of how the model will perform inside your business.

Why the hosted-versus-weights gap matters to buyers

For a business evaluating AI options, the difference between hosted access and released weights changes the decision entirely. Hosted access means the vendor runs the model, and your data leaves your building under the vendor's terms, retention practices, and rate limits. Released weights mean you could, in principle, run the model on your own infrastructure, which changes cost structure, data control, and integration options. A promised release date is useful for planning, but it is not a capability you can rely on today.

There is also a timing wrinkle. The vendor's blog page carries a modification date of October 24 without a timezone, while the independent evaluation is dated October 23. The exact moment of the announcement is not verifiable from the public record, so we treat the launch as current same-day news rather than attaching a precise release time.

What this means for your business

A new hosted model is an option to evaluate, not a system to adopt. The practical question for a document-heavy team is not whether a model scores well on someone else's benchmark, but whether it handles your documents, your templates, and your review standards without leaking records to the wrong people. That is an engineering question, and it is the kind of system Detailed In Design builds.

Concretely, Detailed In Design could build a document intake and reviewed drafting workflow around a hosted model of your choice. Incoming documents would be captured through approved connectors with explicit data permissions and retention rules. Drafts would be source-linked, so every generated paragraph points back to the file it came from. Before any output reaches a customer, a person with the right access reviews and approves it, and the audit history records who approved what and when. The goal is fewer manual handoffs and more consistent review. We make no claim of measured savings or accuracy gains; results depend on your data, your tasks, and your review discipline.

Realistic limitations matter as much as the concept. Using a hosted model requires model access and agreed usage costs, and Detailed In Design has not purchased access to Kimi K3.1 Pro or run test calls against it. Data permissions, retention terms, and integration design must be settled before any record leaves your systems. Task-specific evaluation on your own workflows is required before production use, and human review remains mandatory for customer-facing output. This proposed workflow is a Detailed In Design service concept; it is separate from Moonshot's release and implies no partnership, endorsement, or ownership.

If you want a disciplined evaluation of a new hosted model against your own document and coding tasks, start with a small, bounded test: one workflow, a handful of representative documents, and clear pass-or-fail criteria reviewed by your staff. Detailed In Design can help you design and build that evaluation as part of a custom workflow. See our services at detailedindesign.com.