---
type: "post"
id: "microsoft-foundry-now-hosts-moonshot-ai-s-kimi-3ee6fb"
slug: "microsoft-foundry-now-hosts-moonshot-ai-s-kimi-3ee6fb"
title: "Microsoft Foundry now hosts Moonshot AI's Kimi K3.1 Pro, a 2.8T-parameter open-weight MoE with 1M-token context."
description: "Microsoft says Moonshot AI’s Kimi K3.1 Pro is now available in Microsoft Foundry, with text-and-image understanding and a context window of up to one million tokens."
retrieval_nugget: "Microsoft says Moonshot AI’s Kimi K3.1 Pro is now available in Microsoft Foundry, with text-and-image understanding and a context window of up to one million tokens. The catalog entry supplies per-token prices and gives Foundry users another model to test against their own coding, research and document workloads."
published_at: "2026-11-08"
updated_at: "2026-11-08"
record_date: "2026-11-08"
date_kind: "published_at"
topics: ["models","open-models","multimodal"]
entities: ["Microsoft","Microsoft Foundry","Moonshot AI","Kimi K3.1 Pro"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "Kimi K3.1 Pro is now available through Microsoft Foundry, extending its distribution beyond Moonshot’s own services."
prior_refs: ["https://newruntime.com/posts/moonshot-ai-releases-kimi-k3-1-pro-a-50d51b/"]
source_url: "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/moonshot-ai%E2%80%99s-kimi-k3-1-pro-comes-to-microsoft-foundry/4570292"
source_urls: ["https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/moonshot-ai%E2%80%99s-kimi-k3-1-pro-comes-to-microsoft-foundry/4570292"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:microsoft-foundry-now-hosts-moonshot-ai-s-kimi-3ee6fb"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/microsoft-foundry-now-hosts-moonshot-ai-s-kimi-3ee6fb/","markdown":"https://newruntime.com/posts/microsoft-foundry-now-hosts-moonshot-ai-s-kimi-3ee6fb.md","json":"https://newruntime.com/posts/microsoft-foundry-now-hosts-moonshot-ai-s-kimi-3ee6fb.json"}
markdown_url: "https://newruntime.com/posts/microsoft-foundry-now-hosts-moonshot-ai-s-kimi-3ee6fb.md"
---

# Microsoft Foundry now hosts Moonshot AI's Kimi K3.1 Pro, a 2.8T-parameter open-weight MoE with 1M-token context.

## Retrieval answer

Microsoft says Moonshot AI’s Kimi K3.1 Pro is now available in Microsoft Foundry, with text-and-image understanding and a context window of up to one million tokens. The catalog entry supplies per-token prices and gives Foundry users another model to test against their own coding, research and document workloads.

## Summary
Microsoft says Moonshot AI’s Kimi K3.1 Pro is now available in Microsoft Foundry, with text-and-image understanding and a context window of up to one million tokens. The catalog entry supplies per-token prices and gives Foundry users another model to test against their own coding, research and document workloads.

## What happened
Microsoft’s Foundry blog announced on November 5, 2026, that Moonshot AI’s Kimi K3.1 Pro is coming to Microsoft Foundry. Microsoft describes it as an open-weight model built for agentic work, with text and image understanding and a context window of up to one million tokens. The company presents the arrival as an option for developers building coding, research and document-heavy applications.

Microsoft calls Kimi K3.1 Pro the first model in Moonshot AI’s K3.1 series and says it builds on Kimi K3 with improvements across coding, research and other multi-step tasks. The announcement ties the model’s long context to situations where an agent must retain a large body of relevant material while progressing through work, rather than produce a single isolated answer. It also highlights multimodal use: applications can ask the model to handle images alongside text.

The use cases Microsoft names are practical but illustrative. A coding agent might investigate a change across a large repository, understand related files and work through an engineering task. A research workflow might place lengthy reports and supporting materials in the same task so that the model can examine the material before preparing a synthesis. Applications that work with charts, images or document pages can combine those visual inputs with broader textual context.

Microsoft explicitly cautions that a large context window is not, by itself, a shortcut to a good application. Its announcement says teams should ask whether the model uses the context effectively, completes the task, and does so at a speed and cost that work for the customer. Microsoft says Moonshot AI’s benchmarks show strong results on coding, agentic and multimodal evaluations, while recommending that developers test representative tasks rather than assume those results will transfer directly to an application.

The suggested evaluation is notably framed around completed work rather than a first response. Microsoft recommends comparing quality of completed tasks, reliability with long inputs, places where an agent needs tools, performance on visual material, and cost per successful task. In Microsoft Foundry, Kimi K3.1 Pro can be evaluated alongside other models. Microsoft also says Foundry carries open-weight models from NVIDIA, OpenAI, DeepSeek, MiniMax, Mistral AI, Moonshot AI and Qwen, plus partner offerings such as Fireworks on Foundry.

Microsoft’s pricing table lists Kimi 3.1 Pro at $2.30 per million input tokens, $0.23 per million cached-input tokens, and $11.50 per million output tokens. The announcement directs developers to the Microsoft Foundry model catalog to begin. The immediate change is therefore access through Foundry, with a published pricing structure for a model that Moonshot AI had already released through its own products.

## What the archive adds
New Runtime’s earlier coverage described Kimi K3.1 Pro as a Moonshot AI release with availability across the company’s products: [Moonshot AI releases Kimi K3.1 Pro](https://newruntime.com/posts/moonshot-ai-releases-kimi-k3-1-pro-a-50d51b/). This is a new hosting and pricing fact on that known model story, not a new model release. Microsoft Foundry adds another environment in which teams can evaluate the model alongside alternatives.

## Trend signal
This supports New Runtime’s [owned learning loop shift](https://newruntime.com/shifts/owned-learning-loop/): major platforms are accumulating swappable model options while encouraging customers to judge them with their own rubrics, traces and workloads. Microsoft’s announcement is aligned with that operating model because it emphasizes representative task comparisons and completed-work quality. It does not establish that teams have adopted a particular evaluation process or that Kimi wins those comparisons.

## Why now
The timing follows Moonshot AI’s release of Kimi K3.1 Pro and makes the model available through a second commercial environment. Microsoft frames the addition as part of its continuing investment in open-weight models. For a Foundry customer, this changes the practical question from whether the model exists to whether it delivers a better result, at an acceptable cost, on an existing workflow.

## Why it matters
For teams already building on Foundry, the release reduces the integration friction of testing another long-context, multimodal model. Kimi K3.1 Pro is a plausible candidate for repository-scale coding investigation, document analysis, and workflows that need to relate text to charts, images or scanned pages. The listed rates also permit a first-pass estimate of input, output and cache costs before a pilot begins.

But a one-million-token ceiling should not be treated as a workload outcome. Long context can increase the material an application supplies to a model; it does not demonstrate that the model will identify the right evidence, use tools appropriately, or finish the task reliably. A useful trial holds task definitions and success criteria stable, compares the incumbent model with Kimi, and records the cost and human review needed for each successful completion.

Microsoft’s focus on completed work is the more useful decision rule. A team should select the model that performs acceptably on the work it actually needs done, not the model with the largest headline context window or the most favorable vendor benchmark framing.

## Limits and uncertainty
Microsoft characterizes Moonshot AI’s benchmark results as strong, but those claims remain vendor-reported performance claims rather than independent validation. New Runtime has not independently tested Kimi K3.1 Pro in Microsoft Foundry, including its behavior on very long contexts, coding tasks, visual inputs, tool use, latency or throughput. The published per-token prices are rates, not a total cost of ownership: actual spend will depend on prompt size, output length, cache use, tool calls and retries. The announcement also does not establish that the model will outperform any particular Foundry alternative for a given team.

## Skill to build now
Create a small model-evaluation harness for one real workflow this week. Select a set of representative coding, research or document tasks; define what counts as a completed task before running models; and record quality, human corrections, runtime and token cost. Run the incumbent model and Kimi K3.1 Pro with the same inputs where possible. This turns a catalog addition into a decision backed by completed-work evidence rather than impressions from a demo.

## Reader takeaway
Kimi K3.1 Pro is now a Microsoft Foundry option with a one-million-token context ceiling and published input, cached-input and output rates. Its availability matters most to teams that can run a controlled comparison on real work. Treat the launch as an evaluation opportunity, not proof that long context or vendor benchmarks make it the best model for your stack.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- topic hub: [Open Models](https://newruntime.com/topics/open-models.md): Explore the open-models topic hub.
- topic hub: [Multimodal](https://newruntime.com/topics/multimodal.md): Explore the multimodal topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)