Kimi K2.7 Code in GitHub Copilot: First Open-Weight Model

Update — July 21, 2026: Microsoft is reportedly evaluating Kimi K3 for possible Copilot workloads and preparing Azure availability; Microsoft has not confirmed a deployment or the viral $600M savings estimate. K2.7 remains the confirmed GitHub Copilot precedent. Reported K3 Copilot/Azure testing →

Update — July 20, 2026: Moonshot paused new Kimi K3 subscriptions — GPU capacity crunch. K2.7 Code remains the available-now Kimi path in Copilot while K3 waitlists reopen in batches. K3 subscription pause →

Update (July 9, 2026 — SWE-1.7): Cognition built SWE-1.7 on Kimi K2.7 Code base — 42.3% FrontierCode, available in Devin at 1000 TPS. SWE-1.7 training story →

July 1, 2026: GitHub announced Kimi K2.7 Code is generally available in GitHub Copilot — the first open-weight model selectable in Copilot's model picker. @GitHub reposted the news on X July 2 with a demo video; early community reaction framed it as "the moat keeps getting cheaper" and "open weight in a Microsoft product is like a stray cat becoming a service dog."

This is not a new model release. Moonshot AI shipped Kimi K2.7-Code on June 12, 2026 — the same day the US suspended Fable 5 globally. GitHub's move is distribution: bringing an open-weight 1T-parameter MoE coding model into the same picker where developers already choose Claude, GPT, and Gemini variants.

TL;DR

What GitHub Announced

From the July 1 changelog:

Kimi K2.7 Code, an open-weight model, is now generally available in GitHub Copilot. This is the first open-weight model offered as a selectable option in the Copilot model picker, giving you more choice and a lower-cost option for your coding workflows.

Key details:

• Hosted by GitHub on Microsoft Azure — managed inference, not local weights
• Billed at provider list pricing under usage-based billing
• Early testing: lower-cost option with strong performance comparable to highly popular frontier models (GitHub's wording; no public benchmark table in the changelog)

The @Kimi_Moonshot account co-promoted the launch. For model architecture, benchmarks, and API pricing outside Copilot, see our full guide: Kimi K2.7-Code: 1T-Parameter Open Coding Model.

Why This Matters — Open Weight Inside Microsoft's Stack

GitHub Copilot historically routed requests to closed models — OpenAI GPT family, Anthropic Claude, Google Gemini. Developers who wanted Kimi K2.7 or GLM-5.2 used separate APIs, IDEs, or self-hosted stacks.

Copilot adding Kimi K2.7 changes the default workflow for millions of VS Code users:

@SlopToSignal on X: "open weight AND in the copilot picker is lowkey a big deal — the moat keeps getting cheaper." The Fable 5 ban accelerated open-weight adoption; GitHub just productized that trend inside Copilot.

Important nuance: "Open-weight in Copilot" means the model weights are public — not that inference runs on your laptop. GitHub hosts on Azure. You get model choice and potentially lower cost; you do not get offline/air-gapped inference unless you self-host separately.

How to Enable Kimi K2.7 in GitHub Copilot

Individual plans (Pro, Pro+, Max)

1. Update your editor to a supported version (see table below)
2. Open the Copilot model picker
3. Select Kimi K2.7 Code
4. If missing, wait — GitHub says rollout is gradual

Rollout started with Copilot Pro, Pro+, and Max. GitHub Students on Copilot Pro saw questions on X about whether limits changed — GitHub did not announce a separate student tier change in this changelog; check your plan's premium request quota in account settings.

Supported surfaces and minimum versions

For a broader Copilot vs Claude Code vs Cursor comparison, see Claude Code vs Cursor vs GitHub Copilot — note that guide predates Kimi in the picker; Copilot's model line is now wider.

Copilot Business and Enterprise

Off by default. Plan administrators must:

1. Open Copilot settings for the organization
2. Enable the Kimi K2.7 Code policy
3. Review open-weight models against security, compliance, and data-governance requirements

Until enabled, the model does not appear for anyone in the org. @gsemetfr on X summed up Business-plan frustration: "Damned we want this on business plan!!" — GitHub says Business and Enterprise expansion is coming over the coming weeks.

Pricing and "Lower Cost" in Practice

GitHub states Kimi K2.7 is billed at provider list pricing under usage-based billing. That means:

• Costs depend on your Copilot plan (included premium requests vs overage)
• Kimi may consume fewer premium credits than flagship closed models if list price is lower
• Exact per-request math lives in GitHub Copilot pricing for models and requests

Outside Copilot, Kimi K2.7-Code API pricing from Moonshot is $0.95/M input (cache miss), $4/M output — see the full Kimi pricing table. Copilot routing may differ from direct API calls; treat GitHub's billing docs as authoritative for in-IDE usage.

@Tommi_Lindfors asked on X what "lower cost" means in practice — GitHub has not published a Copilot-specific price sheet alongside the changelog. Run your own A/B on representative tasks (refactors, tests, MCP-style tool loops) and compare premium request consumption against Claude Sonnet or GPT defaults.

Kimi K2.7 vs GLM-5.2 — Why Copilot Picked Kimi First

X replies on GitHub's post repeatedly asked: "Where is GLM 5.2?" (@NoNameTokens, @strugglercss, @RodrigoVie52602)

Both models emerged in the post–Fable 5 ban window:

GLM-5.2 tops BridgeBench Reasoning at 42.8; Kimi K2.7 leads on MCP Mark Verified (81.1% vs Opus 76.4%) in Moonshot's tables. GitHub likely chose Kimi first for agentic/MCP alignment with Copilot's cloud agent and CLI surfaces — but GLM's coding-plan ecosystem (ZCode 3.0 integration) makes it the obvious next ask.

Performance Expectations

GitHub claims early testing shows performance comparable to highly popular frontier models at lower cost. Community skepticism focuses on edge cases — @bullbear_info asked whether Kimi handles 2k-line legacy files without hallucinating types.

From Moonshot's published benchmarks (see Kimi K2.7 guide):

• +21.8% on Kimi Code Bench v2 vs K2.6
• 256K context — relevant for large legacy files
• Thinking mode required — non-thinking requests fall back to K2.6
• Trails Opus 4.8 on some coding benches; beats on MCP Mark Verified

Copilot's managed prompt wrapper may behave differently from raw API calls. Benchmark your repo, not the leaderboard.

Enterprise and Security Considerations

GitHub explicitly recommends administrators review open-weight models before enabling Kimi for Business/Enterprise:

• Data routing: Prompts go to Azure-hosted Kimi inference
• Model lineage: Weights are public — behavior is inspectable, hosting is not self-controlled
• Policy gating: Default-off for orgs is the right default for regulated industries
• Comparison: Closed models offer vendor SLAs; open-weight in Copilot adds a third-party model path inside Microsoft's billing envelope

Teams already evaluating Claude Code vs Copilot for compliance should add open-weight policy review to the checklist when Kimi appears in Business plans.

Timeline — From Fable Ban to Copilot Picker

Same week GitHub also retired Gemini 2.5 Pro and Gemini 3 Flash from Copilot and shipped Copilot CLI improvements — model lineup churn is active. Kimi arrival is the headline for open-weight adoption.

Related Reading

• Kimi K3 subscription pause — K2.7 fallback while K3 waitlisted (July 2026)
• Kimi K3 beta leaks — K2.6 runway to K3 (Jul 2026)
• Kimi K2.7-Code: Full Model Guide — architecture, benchmarks, self-hosting
• GLM-5.2 vs Fable 5 After the Export Ban — the other open-weight response
• Claude Code vs Cursor vs GitHub Copilot — where Copilot fits in your stack
• How to Run GLM-5.2 in Agent Harnesses — until GLM lands in Copilot
• Is Fable 5 Back? Status Hub — why developers sought Kimi in the first place
• GitHub CD-ROM Repo Promotion — same week in GitHub news

Summary

Kimi K2.7 Code in GitHub Copilot is the first open-weight model in the picker — hosted on Azure, rolling out to Pro plans first, off by default for Business and Enterprise. Select it in VS Code 1.127.0+ if your account has received rollout access.

For most developers this is a cost and choice upgrade inside an existing Copilot subscription. For the industry it is a signal: frontier open-weight coding models are now default IDE options, not side projects. GLM-5.2 in Copilot is the obvious next question — GitHub has not answered it yet.

Last updated July 20, 2026. Copilot rollout, policies, and pricing change frequently — verify against GitHub's changelog and your organization's Copilot settings.