翻訳待ち：Kimi K3 from Moonshot AI is now available on Databricks through Unity AI Gateway

AI サービスが一時的に利用できないため、復旧後に翻訳を補完します。

Kimi K3 from Moonshot AI is now available on Databricks through Unity AI Gateway | Databricks Blog Skip to main content Kimi K3, Moonshot AI's most capable open-weight model, is now available on Databricks via Foundation Model API, with full governance through Unity AI Gateway and comprehensive ZDR Kimi K3 is initially hosted by Databricks in the US, with more regions planned. Databricks customers can access Kimi K3 with the same unified API, cost controls, guardrails, and observability they use for every other model on the platform A year ago, the best open-weight models trailed their proprietary counterparts by a wide margin. That gap has narrowed significantly increasing optionality for customers. Today, we're pleased to announce that Kimi K3 from Moonshot AI, the most advanced open-weight model, is available on Databricks. By offering native access to Kimi K3, together with top proprietary models from Anthropic, OpenAI and Google Gemini customers have maximum choice and control, and can build with confidence knowing the models are fully governed by Unity AI Gateway within their secure Databricks perimeter. Kimi K3 is an open-weight model that delivers frontier-level quality on the enterprise tasks that matter most, at a fraction of the cost. For enterprises, this changes the calculus. Customers no longer have to choose between quality and control. Kimi K3 delivers state-of-the-art coding, agentic reasoning, and document understanding while giving you the deployment flexibility, cost efficiency, and customizability that open weights provide, all governed through Unity AI Gateway and comprehensive ZDR (zero data retention) coverage. Kimi K3 ranks #4 with a score of 57 in Artificial Analysis’ Intelligence Index placing it well above average against comparable models. Databricks evaluated Kimi K3 across our internal enterprise benchmarks. The headline: Kimi K3 consistently performs at the level of leading proprietary models, often at 50–70% lower cost per task. Context, control, choice, and cost Enterprise AI isn't just about picking the best model. It's about connecting that model to your data, governing how it's used, having the freedom to evolve your model strategy, and doing it all at a cost that scales. On Databricks, this comes down to four principles: context, control, choice, and cost. Context: AI is only as good as the data it reasons over. On Databricks, Kimi K3 runs directly against your governed enterprise data through Unity Catalog. Your agents can query your lakehouse, access your feature stores, and ground their responses in your organization's actual knowledge, not generic training data. This is what turns a capable model into a capable enterprise model. Control: Frontier-quality models need frontier-quality governance. Unity AI Gateway gives platform teams a single control plane across every model: fine-grained access permissions, safety guardrails on every request and response, complete audit trails logged to Unity Catalog, and spend controls that prevent runaway costs. Since Databricks is hosting Kim K3 open weights, your data remains in your lakehouse, within Databricks' security perimeter. The same governance policies you apply to proprietary models apply identically to Kimi K3. Invoke Kimi K3 via REST API: Choice: The model landscape is evolving faster than any single vendor can keep up with. Databricks gives you the freedom to use the best model for each task without lock-in, across both open weight models such as Kimi 3 and proprietary models such as Anthropic Claude, OpenAI GPT, or Google Gemini. With Databricks intelligent routing across your entire model portfolio, users can easily combine Kimi K3 for cost-efficient coding with other models of different strengths, all through a single, standardized API. When a newer model emerges next quarter, you swap it in without rewriting a single line of application code. Cost: Frontier quality no longer requires frontier pricing. Kimi K3 delivers 50–72% lower cost-per-task than comparable proprietary models, and Unity AI Gateway's Spend Controls let platform teams set budgets per user, team, workspace, and account with alerts before limits are hit. The Cost Analytics dashboard provides full visibility into spend by model, provider, and use case. The result: agentic workloads that were previously cost-prohibitive (coding agents running continuously, document pipelines processing millions of pages) become economically viable at scale. Kimi K3 is a powerful proof point of this philosophy: an open-weight model that matches proprietary quality, deployed with enterprise governance, at a fraction of the cost, alongside every other model your organization uses — all on one platform. Get started Kimi K3 is available now, with US-hosting on Databricks Foundation Model API with native access for AWS and GCP workspaces and ADI access for Azure Databricks. Try Kimi K3 in AI Playground: Explore the model's capabilities directly in your Databricks workspace with zero setup. Configure Unity AI Gateway: Set up governance, cost controls, and model routing to add Kimi K3 to your existing model portfolio. Build Kimi-powered agents with Agent Bricks: Analyze your data, automate complex work, evaluate with custom judges, and deploy at scale. Contact your account team: For capacity requests, regional availability, or help designing your model routing strategy. Get the latest posts in your inbox Subscribe to our blog and get the latest posts delivered to your inbox. Sign up View all blogs