← Back to all sparks
GitHub Copilot logo

GitHub Copilot

AI-ASSISTANTS
Velocity10.0

GitHub's AI pair-programming assistant from Microsoft / GitHub.

Copilot's week is model housekeeping and cost accounting, not new capability.

model-lifecyclecost-accountingusage-reportingenterprise-controlscode-review
Current state
Copilot is in a consolidation stretch: the six most recent posts are a small-tier model swap, a per-model token breakdown in the usage report, an ROI section in the impact dashboard, and chat UI cleanup. The only model news is MAI-Code-1.1-Flash replacing MAI-Code-1-Flash, with the older model dated for removal on September 10. Nothing in this window expands what an agent can do; it changes what a buyer can see and what an admin can control.
Where it's heading
The center of gravity has moved from shipping capability to proving and governing it. Usage reporting now breaks AI credits down per model, the impact dashboard converts spend into pull request output, and the usage metrics API counts third-party agent app activity — three surfaces aimed squarely at whoever approves the invoice. Model news has shifted in kind too: from frontier launches like GPT-5.6 and Claude Sonnet 5 earlier this summer to lifecycle management, where a deprecation ships with a date and a named successor.
Prediction
Expect the next stretch of posts to stay on the billing and administration surface — more usage-report dimensions and enterprise controls — with model entries arriving as scheduled deprecations rather than new families.

Recent moves

  1. 2h ago

    Upcoming deprecation of MAI-Code-1-Flash

    The retirement notice that pairs with the 1.1-Flash rollout: MAI-Code-1-Flash leaves every Copilot surface on September 10, with the new model named as the replacement. Copilot now publishes model deprecations with a date and a migration path rather than quietly rotating the picker.

    View source ↗
  2. 3h ago

    MAI-Code-1.1-Flash available in GitHub Copilot

    A point upgrade to Microsoft's small-tier coding model, adding image understanding and general coding-quality gains. Unlike the GPT-5.6, Kimi K2.7, and Claude Sonnet 5 additions earlier this summer, this refreshes a model already in the picker rather than adding a new family to it.

    View source ↗
  3. 7h ago

    Per-model token breakdown in the usage report

    AI credits now itemize into input, output, and cache tokens for each model. This is the reporting counterpart to the impact dashboard's ROI work — Copilot spend is becoming something an admin can attribute to a model rather than read as one line.

    View source ↗
  4. 1d ago

    Copilot on web expands conversation controls

    Easier access to recent conversations and a minimizable chat panel on github.com. Ordinary interface work, and the only entry in this window that touches the day-to-day chat surface rather than the reporting layer around it.

    View source ↗
  5. 3d ago

    Copilot desktop, CLI, and VS Code gain session resume

    The weekly roll-up across the desktop app, CLI, and VS Code, themed on resuming and organizing work without losing context. The digest format keeps each individual change small; the signal is that all three clients now move on a shared weekly cadence.

    View source ↗
  6. 4d ago

    Copilot impact dashboard adds a return on investment section

    The impact dashboard now ties what a team spends on Copilot to the pull request output it gets back. It is the clearest statement in this window that the current cycle is about defending the line item rather than extending the tool.

    View source ↗