Ollama has announced that Kimi K3.1 Pro will begin rolling out on Ollama Cloud shortly, with US-based hosting and zero data retention. The November 4 announcement brings Moonshot AI’s latest flagship into focus for developers who want hosted access through Ollama’s ecosystem. Ollama’s announcement
The important distinction is where the model runs and what happens to the information sent to it. Zero data retention concerns the handling of request content. Cloud inference still involves sending that content to a remote service.
What Kimi K3.1 Pro brings
Moonshot describes Kimi K3.1 Pro as the first model in its K3.1 series, with a one-million-token context window and improvements in coding, 3D creation, financial research and agentic work. The company says it builds on Kimi K3 through substantially expanded post-training. Moonshot’s model announcement
Those are the model developer’s capability claims. Their practical value depends on the task, the tools available around the model and the quality of the result.
For example, a useful coding evaluation should ask whether the assistant produces a correct, maintainable change in an existing repository. A financial-document evaluation should check whether calculations and conclusions can be traced to the supplied evidence.
A large context window provides room for more material. It does not establish that every relevant detail will be used correctly.
What Ollama means by zero data retention
Ollama’s FAQ says cloud prompts and responses are processed to provide the service, without storing or logging that content or using it for model training. It distinguishes that content from basic account information and limited usage metadata. Ollama FAQ
Its privacy policy provides additional detail: cloud content is processed transiently and is not stored beyond the time required to fulfil the request. Account information, billing records, support communications and operational metadata have separate retention rules. Ollama Privacy Policy
That distinction matters when evaluating a privacy claim. The commitment concerns prompt and response content; running a paid service still involves managing accounts and transactions.
Three privacy questions worth separating
| Question | What Ollama states | Practical meaning |
|---|---|---|
| Are cloud prompts and responses retained? | The FAQ says they are not stored or logged; the policy describes transient processing to fulfil the request. | The service processes content while answering. |
| Are inputs and outputs used for training? | Ollama says they are not. | Training policy and retention policy are separate commitments. |
| Does the service retain other information? | Account, billing, support and usage-related information follow separate rules. | Zero content retention does not mean the entire service keeps no records. |
Sources: Ollama FAQ and Privacy Policy.
A team should also examine its own application. A developer tool could save conversation history, an internal gateway could record requests, or a workflow could copy an answer into a document. The provider’s handling policy does not determine how those other components are configured.
For a practical deployment, map the route from the application to the model and back. Identify which component can retain content at each stage.
US hosting and local execution answer different needs
Ollama’s Kimi K3.1 Pro announcement specifies US-based hosting. That is a material detail for anyone whose project requires processing in a particular region. Announcement
Ollama supports both local and cloud workflows. Its cloud documentation explains that hosted models do not need to be downloaded, and that applications can call the service directly without installing Ollama. Cloud documentation
For developers, this can make it easier to try a demanding model without provisioning the inference infrastructure themselves. The trade-off is that access depends on the remote service and its terms.
Local execution meets a different requirement: keeping model inference on infrastructure you control. The fact that a request starts in a local terminal or editor does not establish that the model itself runs locally.
Readers considering the commercial side can also consult our guide to Ollama Cloud pay-as-you-go access and Pro.
The rollout announcement is not a universal access guarantee
Ollama’s wording is forward-looking: the model is “rolling out on Ollama Cloud shortly.” It does not specify a completed rollout, an exact activation time or access for every account. Original announcement
For API integrations, Ollama documents a way to list its cloud models:
curl https://ollama.com/api/tagsUse the model identifier returned by the service for API requests; naming conventions in the application or CLI can differ. Ollama’s model-listing guidance
Before changing an existing workflow, confirm that the intended model is available to your account and that its current rate and supported features match your needs. Moonshot’s own API and an Ollama-hosted route are separate service choices.
A model announcement also does not bundle every feature of the developer’s own applications. Access to a model and access to a complete research, coding or design workspace should be evaluated separately.
How to assess it on useful work
Our suggested first trial is a small task with a clear result and an affordable way to verify it. For a coding assistant, that could be fixing one reproducible issue with existing tests. For document work, it could be answering a defined set of questions with references to the supplied files.
Keep the task and acceptance criteria consistent when comparing models. Record completion quality, elapsed time, usage cost and the corrections required. Include failed attempts when assessing value.
For a workflow using private material, start with a representative sample that is appropriate for the chosen environment. Check the application’s own history and logging settings alongside the provider’s content-handling policy.
Our guide to what you can actually delegate to AI agents explains why a checkable outcome is a better starting point than a broad instruction to handle an entire process.
What the announcement changes
Our assessment is that the combination could make Kimi K3.1 Pro easier to evaluate for teams already using Ollama. The significance lies in the available route to the model as well as its capabilities.
The decision should rest on three things: whether the model performs the required work, whether the hosting and content-handling arrangements fit the project, and whether the total cost is justified. The rollout announcement is a reason to investigate those questions.
Sources
- Ollama — Kimi K3.1 Pro rollout announcement, November 4, 2026.
- Moonshot AI — Introducing Kimi K3.1 Pro.
- Ollama — FAQ.
- Ollama — Privacy Policy.
- Ollama — Cloud documentation.
Information checked on November 4, 2026.