Reading · Developers Digest

This site does not allow itself to be embedded. You can still read it on the original site — the toolbar below keeps your place in the directory.

GitHub made Kimi K3 generally available in Copilot on August 6 at $3/$15 per million tokens, hosted on Fireworks AI. It is off by default for Business and Enterprise, the rollout was paused mid-day by a GitHub Actions incident, and it changes the price/quality calculus in the model picker.

Kimi K3, Moonshot AI's 2.8-trillion-parameter open-weights model, is now generally available in GitHub Copilot. GitHub [announced the rollout on August 6](https://github.blog/changelog/2026-08-06-kimi-k3-is-now-available-in-github-copilot), with the model hosted by GitHub on Fireworks AI and billed at provider list pricing under usage-based billing: $3 per million input tokens, $15 per million output tokens, and $0.30 per million cached input tokens. It has been a bumpy day in the model picker: GitHub paused the rollout mid-morning while mitigating a GitHub Actions incident, then resumed it the same day. For Copilot Business and Copilot Enterprise, Kimi K3 ships off by default - an administrator has to enable the Kimi K3 policy before anyone in the org can select it. ## What actually changed Kimi K3 joins the Copilot model catalog at GA status across all plans: Copilot Pro, Pro+, Max, Business, and Enterprise. It appears in the model picker in Visual Studio Code, Visual Studio, Copilot CLI, the Copilot cloud agent, the Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains, Xcode, and Eclipse. The official [models and pricing documentation](https://docs.github.com/copilot/reference/copilot-billing/models-and-pricing) already lists it: | Model | Input /1M | Cached input /1M | Output /1M | Status | | --- | --- | --- | --- | --- | | Kimi K3 | $3.00 | $0.30 | $15.00 | GA | | Kimi K2.7 Code | $0.95 | $0.19 | $4.00 | GA | | Claude Opus 5 | $5.00 | $0.50 | $25.00 | GA | | GPT-5.6 Luna | $0.20 | $0.02 | $1.20 | GA | | Gemini 3.6 Flash | $1.50 | $0.15 | $7.50 | GA | Moonshot now has two price points in Copilot: the K2.7 Code budget tier at $0.95/$4 and K3 at $3/$15. The pricing matches what Moonshot charges on its own API and what OpenRouter resells K3 at, so there is no Copilot-specific markup on the token rate. ## The rollout pause matters more than it looks The changelog carries two editor's notes dated August 6. The first says the rollout was temporarily paused while GitHub mitigated an incident with GitHub Actions, and that pricing documentation would follow with the K3 rates. The second confirms the rollout resumed and the model is billed at provider list pricing. That is a rare public look at how model rollouts are operated. The pause and resume happened inside one day, which means GitHub's model catalog, billing pipeline, and Actions infrastructure are coupled tightly enough that an incident in one can gate the other. Teams that depend on a specific model being available should not treat Copilot's catalog as static - this is the same lesson as the [Gemini 2.5 Pro and Gemini 3 Flash deprecation](/blog/github-copilot-gemini-models-deprecated-2026) from a week earlier, just from the addition side instead of the removal side. ## Why Kimi K3 in Copilot matters for developers First, access. Kimi K3's open weights (1.56 TB on Hugging Face) make it the largest self-hostable model ever, but hosting 2.8T parameters is a real operations problem. Copilot gives a turnkey route: pick it in the menu, no infrastructure. That is a genuinely new access tier, and we walked through every other route in our [Kimi K3 access comparison](/blog/where-to-access-kimi-k3-2026) - Copilot is the first that costs zero setup. Second, the price/quality position. At $3/$15, K3 sits between the budget tier and the frontier tier in the Copilot catalog: 3x the input price of K2.7 Code, 1.7x the price of Gemini 3.6 Flash, but about 40% cheaper than Claude Opus 5 on input and 40% cheaper on output. Run the standard agent math on it: a 40-turn tool-calling task at 2M cumulative input tokens (75% cache-hittable) and 60K output costs about $2.85 on K3, versus roughly $4.75 on Claude Opus 5, $1.00 on K2.7 Code, and $0.20 on GPT-5.6 Luna at the same rates. K3 is not the cheap tier - it is the "frontier-ish capability without the frontier price" tier, the same positioning it holds in the [open-weights market](/blog/glm-5-2-cost-math-open-weights-coding-models). Third, the enterprise signal. Copilot is quietly becoming a multi-vendor marketplace with policy controls: the [enterprise model policy targeting](/blog/github-copilot-enterprise-team-model-policy-2026) in public preview, the [GitHub Models retirement](/blog/github-models-retired-2026) that ended the neutral API surface, and now third-party model GAs on a rolling cadence. Kimi K3 being off by default for Business and Enterprise is the new normal for external models: the admin, not the individual developer, decides. For individual devs on Pro and Max plans, the model picker just got a serious new option. ## How to try it If you are on Copilot Pro or Max, open the model picker in VS Code or Copilot CLI and select Kimi K3; the changelog says rollout is gradual, so it may take hours to reach every account. On Business or Enterprise, ask an admin to enable the Kimi K3 policy in Copilot settings first - the model stays invisible until the policy flips. For API users, K3 remains available on Moonshot's platform and through the providers listed in our [access guide](/blog/where-to-access-kimi-k3-2026); Copilot is an additional surface, not a replacement for the API. One caveat worth stating: usage-based billing in Copilot converts tokens to AI credits (1 credit = $0.01), and allowances vary by plan, so your effective cost depends on whether the task lands inside or outside your plan's included credits. The per-token rates above are the over-allowance price. ## FAQ ### Is Kimi K3 free in GitHub Copilot? No. It is billed at provider list pricing under usage-based billing: $3 per 1M input, $0.30 per 1M cached input, $15 per 1M output. Usage inside your plan's included AI credits allowance is covered by the subscription. ### Is Kimi K3 available in Copilot Business and Enterprise? Yes, but off by default. A plan administrator must enable the Kimi K3 model policy in Copilot settings before any org member can select it. ### What happened with the Kimi K3 rollout on August 6? GitHub paused the rollout temporarily to mitigate a GitHub Actions incident, then resumed it the same day. The changelog was updated twice to reflect the pause and the resume. ### How does Kimi K3 in Copilot compare to Claude Opus 5? K3 is cheaper per token ($3/$15 vs $5/$25) but sits below Opus 5 on frontier benchmarks. It is the mid-price option: more capable than the budget tier, less expensive than the frontier tier. ## Sources - [GitHub Changelog: Kimi K3 is now available in GitHub Copilot](https://github.blog/changelog/2026-08-06-kimi-k3-is-now-available-in-github-copilot) - announcement, rollout status, and pricing notes - [Models and pricing for GitHub Copilot](https://docs.github.com/copilot/reference/copilot-billing/models-and-pricing) - verified per-token rates for Kimi K3 and the rest of the catalog - [Kimi K3 model card on Hugging Face](https://huggingface.co/moonshotai/Kimi-K3) - open weights and license terms ## Continue Reading - [Where to Access Kimi K3: Every Provider Compared](/blog/where-to-access-kimi-k3-2026) - the full access map, now with Copilot as a zero-setup route - [Kimi K3 vs Kimi K2.7](/blog/kimi-k3-vs-k2-7) - how the two Moonshot tiers differ on architecture and price - [Kimi K3 Developer Guide](/blog/kimi-k3-developer-guide) - architecture, capabilities, and API usage - [Gemini 2.5 Pro and Gemini 3 Flash Deprecated in Copilot](/blog/github-copilot-gemini-models-deprecated-2026) - the other side of Copilot's model churn - [AI Coding Tools Pricing 2026](/blog/ai-coding-tools-pricing-2026) - the full subscription and API pricing landscape

Kimi K3 Is GA in GitHub Copilot: Pricing, Rollout, and What It Means for Model Choice

Comments