GitHub shipped its weekly Copilot update on August 10 with two changes that reach everyday coding workflows: a new reasoning model for paying subscribers and a portable plugin format that works across editors and the terminal. The releases, documented in the official GitHub Changelog, roll out incrementally rather than in a single big launch, so developers may see features land in different surfaces over the coming days GitHub Changelog.
The headline model addition is Kimi K3, which is reaching Copilot Pro, Pro+, Max, Business, and Enterprise plans according to the changelog entry GitHub Changelog. Alongside it, MAI-Code-1.1-Flash is rolling out inside Copilot with native image understanding plus improvements to coding quality, instruction following, tool use, and raw performance. Neither model replaces the Claude and Gemini options already in the picker; they widen the menu that users can switch between per turn.
The more structural change is Agent Plugins 1.0, now generally available in VS Code, the Copilot CLI, the Copilot SDK, and the standalone Copilot app, according to the changelog GitHub Changelog. The pitch is simple: build a plugin once and reuse it across those compatible agent tools instead of reimplementing behavior per environment. The GitHub Copilot app page frames the app as a desktop surface for agent-driven development on macOS, Windows, and Linux, where installed plugins can now be managed in one place, version-checked, and updated individually or all at once from Customize or Settings GitHub Copilot app.
On the command line, the update adds practical controls for people running autonomous agents, as listed in the changelog GitHub Changelog. A new /tasks view lists subagents and their work, and you can queue prompts, shell commands, and supported slash commands while an agent turn is already running. Combining –plan with –mode autopilot in headless mode (-p) lets the agent produce a plan and then implement it without step-by-step approval, while /rewind restores Copilot’s changes without touching git or discarding your own edits, as detailed on the GitHub Copilot CLI page GitHub Copilot CLI.
JetBrains users get two quality-of-life upgrades documented in the changelog GitHub Changelog: Copilot memory carries project context across agent chat sessions so preferences don’t have to be repeated, and Ollama now works as a bring-your-own-key model provider for local inference. VS Code 1.133, shipping in the same window per the changelog GitHub Changelog, lets you switch between Claude BYOK and built-in Copilot models for each new turn, pin a prompt while scrolling a long chat, and see HTML changes in the integrated browser without a manual refresh.
For readers tracking Copilot’s model roster, the Kimi K3 addition follows other recent reasoning-model integrations such as xAI’s Grok 4.6 in the same product GitHub Copilot adds xAI’s Grok 4.6 reasoning model. The takeaway for teams is that Copilot is consolidating its agent tooling into a single plugin format while widening model choice, which matters if you work across VS Code, JetBrains, and the terminal and want consistent behavior everywhere.
