Moonshot AI's open-weight coding model — now the first open-weight option in the GitHub Copilot model picker.
Last verified · 2026-07-02 · by Moe Ameen
Kimi K2.7 Code is an open-weight coding model from Moonshot AI, the Beijing lab behind the Kimi series. Moonshot released it on June 12, 2026, publishing the full weights to Hugging Face under a Modified MIT license that permits commercial use and self-hosting. On July 1, 2026, GitHub made it generally available in the Copilot model picker and called it the first open-weight model offered as a selectable option there — positioning it as a lower-cost choice next to Copilot's closed models from OpenAI, Anthropic, and Google.
Under the hood it is a Mixture-of-Experts model: roughly one trillion total parameters, but only about 32 billion active for any given token, which is what keeps inference cheaper than a dense model of similar scale. It has a 256K-token context window and a forced "thinking" (reasoning) mode that cannot be turned off. Moonshot built it specifically for agentic coding — writing, editing, and reasoning over code across a whole project rather than autocompleting a single line.
Moonshot's launch numbers show gains over the prior K2.6 on the lab's internal coding benchmarks, and it cites a strong result on an agent/tool-use suite. As of release, independent third-party results on the common public leaderboards were still limited, so treat the headline benchmarks as first-party until outside evaluations land. In Copilot, the model is hosted by GitHub on Microsoft Azure and billed at provider list pricing under usage-based billing; Moonshot's own API lists roughly $0.95 per million input tokens and $4 per million output tokens.
In Copilot it began rolling out to Pro, Pro+, and Max plans, with Business and Enterprise following — and it is off by default for those org plans until an administrator enables the policy. It is selectable across VS Code 1.127.0+, Visual Studio 17.14.6+, JetBrains 1.9.1-251+, Xcode, Eclipse, the Copilot CLI, github.com, and GitHub Mobile.
Kimi K2.7 Code helps you build the thing. It does nothing to help you tell anyone about it — and for most founders and creators, the marketing is the harder half. That gap is exactly where Kompozy sits. You shipped a feature, a landing page, a whole product with Copilot and Kimi; now you have a launch, a changelog, a demo, a "how I built this" story. Paste any of that into Kompozy and it becomes a build-in-public content unit: a captioned Persona Short or Marketing Short walking through the feature, a Carousel breaking down the build, Quote Graphics pulling the best lines, native Text Posts for X and LinkedIn, a Blog Article for SEO, and an Email Newsletter to your list — all written in your voice through one Persona Brief, then scheduled and published across nine platforms from a single queue.
The pairing is clean because the two tools own opposite ends. Kimi K2.7 Code is a reasoning engine for source code inside an editor; it has no notion of captions, aspect ratios, brand voice, a posting schedule, or a persona's face. Kompozy is the generation-and-publishing engine that produces net-new content Kimi can't touch — avatar and persona video, clipped shorts from a long demo recording, brand-exact carousels and images, blogs and newsletters — and then handles the fan-out and the publish. Ship with Kimi; market the ship with Kompozy.
It is an open-weight coding model from Moonshot AI, released June 12, 2026 under a Modified MIT license. It is a Mixture-of-Experts model with roughly one trillion total parameters and about 32 billion active per token, a 256K-token context window, and a forced reasoning mode. It is built for agentic coding — writing and editing code across a project rather than single-line autocomplete.
Yes. GitHub made it generally available in the Copilot model picker on July 1, 2026 and called it the first open-weight model offered there. It began rolling out to Copilot Pro, Pro+, and Max plans, with Business and Enterprise following — and it is off by default for those org plans until an admin enables the policy.
In Copilot, GitHub hosts it on Microsoft Azure and bills at provider list pricing under usage-based billing, so cost follows GitHub's rates. Moonshot's direct API lists roughly $0.95 per million input tokens and $4 per million output tokens — cheaper than most frontier closed models, which is why GitHub frames it as a lower-cost option.
No. It is a coding model that writes and reasons over source code in an editor. It does not produce captioned video, brand-consistent images, carousels, or scheduled social posts. To turn what you build into on-brand content, pair it with a content engine like Kompozy, which generates across 18 formats and publishes to nine social platforms plus blog and email.
Yes. Moonshot published the full weights to Hugging Face under a Modified MIT license that allows commercial use and self-hosting, with day-one support noted for inference stacks like vLLM and SGLang. The GitHub Copilot integration is a separate hosted path — that runs on Microsoft Azure under GitHub's billing.