Product1 publisher3 min readPublished
Copilot drops the flagship model, and the build record does not follow
GitHub's latest changelog makes per-turn model switching normal across Copilot surfaces. It does not add a field anywhere recording which model wrote which line.
The Product Desk · Product desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- GitHub published its latest round of Copilot updates in its weekly changelog on Aug. 13, covering the week of Aug. 10.
- Kimi K3 is rolling out across Copilot Pro, Pro+, Max, Business, and Enterprise plans.
- MAI-Code-1.1-Flash arrives alongside Kimi K3, with native image understanding and what GitHub describes as improvements in coding quality, instruction-following, tool use, and performance.
- Neither new model replaces anything; they sit next to the models already available in Copilot.
- In Visual Studio Code 1.133, developers can switch models within a Claude session, moving between Claude BYOK and built-in Copilot models on a per-turn basis, without restarting the session or losing context.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
GitHub's Aug. 13 Copilot changelog, covering the week of Aug. 10, added two more models and turned mid-session model switching into a per-turn operation in Visual Studio Code 1.133 [1] [2] [3] [5]. The practical question for enterprise teams has moved: not which model to standardise on, but how to establish which model produced a given line, because Copilot still does not record that [6].
The two additions follow the pattern. Kimi K3 is rolling out across Copilot Pro, Pro+, Max, Business, and Enterprise, which is every named plan tier in the release [2] [18]. MAI-Code-1.1-Flash arrives beside it with native image understanding and what GitHub describes as improvements in coding quality, instruction-following, tool use, and performance [3]. Neither replaces anything already available [4]. That is the design decision worth noting. A vendor that believed in one flagship model would deprecate something.
VS Code 1.133 is where the shift becomes structural: a developer can move between Claude BYOK and built-in Copilot models within a single Claude session, per turn, without restarting or losing context [5]. Mitch Ashley, VP and practice lead for software lifecycle engineering and AI-native software engineering at The Futurum Group, argues the convenience carries a cost. "Developers already run two or three models and pick by task," Ashley said. "Per-turn switching removes the restart and the record of which model wrote which line. The next engineer reviewing that code cannot reconstruct it" [7]. Model choice, in his framing, "is now a runtime dependency with no field in the build record," and by the time something fails in production the choice is invisible [8].
The rest of the release is plumbing, and mostly good plumbing. Agent Plugins 1.0 is generally available and behaves identically across VS Code, Copilot CLI, the GitHub Copilot SDK, and the Copilot app, four surfaces that previously required separate builds of the same internal tooling [9] [19]. The Copilot app gained matching plugin management, including version visibility and individual or bulk updates from Settings [10], plus side chat for answering an agent's clarifying question without derailing the main thread [11].
Copilot CLI took the largest share. A /tasks command manages subagents and reports their status in the terminal [12]. Prompts, commands, and slash commands can be queued while an agent is mid-task [13]. The --plan and --mode autopilot flags can now be combined in headless mode, automating planning and implementation in one pass for CI pipelines and scheduled jobs with nobody watching [14]. And /rewind undoes Copilot's changes without depending on git [15], which is useful in a messy repo and also a second place where a change happens outside version control.
On JetBrains, Copilot Memory carries context between chat sessions [16], and Ollama is supported as a bring-your-own-key provider, extending model choice to models running locally on a developer's machine [17].
Two things to watch. First, whether GitHub ships any per-turn attribution metadata, since headless autopilot in CI [14] plus unlogged switching [6] means an unattended pipeline can produce code no human can trace to a model. Second, whether local Ollama models [17] and BYOK sessions [5] get treated as auditable inputs or stay invisible to whoever reviews the diff.