← Back to feed

GitHub Copilot has offered model choice since late 2025, but every selectable option has been proprietary until now. Kimi K2.7 Code, Moonshot AI’s open-weight coding model, is generally available in Copilot’s model picker as of July 1 — the first open-weight model to reach the standard roster alongside Claude Fable 5, GPT-5.5, and Gemini 3.5 Flash.

The distinction matters beyond the label. Open-weight means organisations can inspect the model weights, fine-tune for domain-specific workflows, and — for teams not on GitHub’s infrastructure — self-host at reduced cost. The model enters Copilot as a cloud-hosted option, not a self-hosted integration, but its inclusion signals that Microsoft’s enterprise AI platform is willing to compete on price and transparency, not just capability.

Availability and Rollout

Kimi K2.7 Code is rolling out to Copilot Pro, Pro+, and Max plan holders via VS Code model picker (version 1.127.0 or later), Visual Studio (17.14.6 or later), github.com, GitHub Mobile (iOS and Android), and JetBrains (version 1.9.1-251 or later). Rollout is gradual.

For Copilot Business and Enterprise plans, the model is off by default. Plan administrators must enable the Kimi K2.7 Code policy in Copilot settings before any user in the organisation can select it.

What Kimi K2.7 Code Brings

Kimi K2.7 Code launched in open-source on June 10. The model ships 30% fewer thinking tokens per task than K2.6 and claims an 81% verified score on MCP Mark — Moonshot’s agentic tool-calling benchmark. API pricing is $4/M output tokens, undercutting Opus 4.8 ($25/M) and Fable 5 by a wide margin. On Artificial Analysis’s coding agent index, Kimi K2.7 Code sits in the competitive mid-tier at a cost advantage over frontier proprietary models.

The open-weight release was under a permissive license, and Moonshot has published full model weights. GitHub is hosting the model in its own inference infrastructure, not routing through Moonshot’s API.

The Broader Copilot Shift

GitHub’s July 1 changelog also added Copilot CLI automatic task-based model routing — the CLI now selects the model based on task type rather than defaulting to the most capable option. Claude Sonnet 5 reached general availability in Copilot on June 30, and the Copilot Agent became available in JetBrains AI Assistant simultaneously.

The cumulative effect of these changes is that Copilot is moving from a single-model product toward a routing layer — selecting among several models based on task complexity, cost, and capability. Kimi K2.7 Code’s inclusion suggests the picker is no longer curated purely for maximum benchmark performance. Cost-per-task and licence transparency are now explicit selection criteria.

What This Means for the Coding Assistant Market

Cursor, GitHub Copilot, and AWS Kiro have been competing for developer mindshare with increasingly overlapping model rosters. Until now, all three platforms sourced exclusively from proprietary frontier APIs. Kimi K2.7 Code breaks that pattern for GitHub. If the model delivers developer-acceptable results at lower cost per task — a reasonable assumption given its position on coding leaderboards — it will pressure competing platforms to either match the open-weight option or compete on model quality at equivalent price points.

The model is available now for eligible plans. GitHub says expansion to Business and Enterprise surfaces will follow in the coming weeks.