GitHub Expands Copilot Model Switching and Agent Tools
GitHub's August 13 Copilot changelog adds Kimi K3 and MAI-Code-1.1-Flash, per-turn model switching in VS Code Claude sessions, and new agent tooling, according to DevOps.com. The update also extends Copilot CLI capabilities and introduces Agent Plugins 1.0 across Copilot surfaces. DevOps.com frames the release around model choice, while raising governance questions about tracing model contributions to generated code.
GitHub's August 13 Copilot changelog adds Kimi K3 and MAI-Code-1.1-Flash to its coding-assistant model lineup, enables per-turn model switching in Visual Studio Code Claude sessions, and expands Copilot CLI and plugin capabilities, according to DevOps.com.
Kimi K3 is rolling out to Copilot Pro, Pro+, Max, Business, and Enterprise subscribers, DevOps.com reports. MAI-Code-1.1-Flash arrives with image understanding and GitHub-described improvements in coding quality, instruction following, tool use, and performance. The report characterizes the additions as expanding the available model menu rather than replacing existing models.
Switching models without restarting sessions
VS Code 1.133 lets developers change models within a Claude session on a per-turn basis, including switching between Claude bring-your-own-key configurations and built-in Copilot models without restarting the session or discarding context, according to DevOps.com.
The Copilot CLI update includes subagent management, queued prompts, headless autopilot, and a /rewind command, the publication reports. Separately, GitHub announced in an April changelog that Copilot CLI's generally available auto model-selection mode routes requests among models based on a user's plan and administrator policies, while allowing users to select a specific model instead. GitHub listed GPT-5.4, GPT-5.3-Codex, Sonnet 4.6, and Haiku 4.5 as examples of models available to auto, noting that its routing set can change over time.
Shared plugins and governance questions
DevOps.com also reports that Agent Plugins 1.0 creates a shared plugin system across Copilot surfaces. JetBrains support adds Copilot Memory and Ollama support, according to the report.
Mitch Ashley, VP and practice lead for software lifecycle engineering and AI-native software engineering at The Futurum Group, told DevOps.com that per-turn switching can complicate review and provenance. He said, "Developers already run two or three models and pick by task," and argued that the absence of a record identifying the model behind a particular code change makes later reconstruction harder.
That concern reflects a broader engineering pattern: teams using multiple coding models often need provenance data alongside source control, test results, and deployment metadata when model output is material to implementation decisions. The reported features lower the friction of changing models during a task, but DevOps.com notes that Copilot does not record which model contributed to which code.
Key Points
- 1GitHub added Kimi K3 and MAI-Code-1.1-Flash, broadening Copilot's available coding-model options; Kimi K3 is rolling out across Pro, Pro+, Max, Business, and Enterprise plans.
- 2Per-turn switching in VS Code retains session context, reducing workflow friction when developers choose different models for distinct tasks.
- 3Multi-model coding workflows commonly intensify provenance needs because source-control records may not identify the model behind generated changes.
Scoring Rationale
The release adds practical multi-model workflow controls across IDE and CLI environments. Its greatest relevance is for engineering teams managing model selection, agent workflows, and code provenance rather than for model-training practitioners.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
