Source: GitHub — 2026-08-11
Summary
Microsoft's MAI-Code-1.1-Flash, a small-tier coding model, is now live in GitHub Copilot with native vision support — it can reason over screenshots, diagrams, and UI mockups alongside raw code, not just text. Versus its predecessor MAI-Code-1-Flash, GitHub reports a 73% lower list price, 25% faster token streaming, 25% fewer tokens consumed per task, 22% better CLI-task performance, and 15% better .NET-task performance. It's rolling out across every Copilot tier and surface — Free and Student get it via auto-model-selection, Pro/Business/Enterprise can pick it manually — spanning VS Code, Visual Studio, JetBrains, Eclipse, Xcode, GitHub Mobile, Copilot CLI, and GitHub's cloud coding agent.
Key Takeaways
- Native vision support is the headline capability shift — the model can now take a screenshot of a broken UI or a hand-drawn diagram as direct input alongside code, rather than requiring everything to be described in text first.
- The efficiency gains stack together: 25% fewer tokens per task combined with 25% faster streaming means both lower cost per task and a faster perceived response, not just one or the other.
- A 73% list-price cut on a model that's also faster and uses fewer tokens is a compounding cost reduction, not a simple discount — the per-task cost drop is larger than the headline price cut alone suggests.
- Task-specific gains (22% better on CLI tasks, 15% better on .NET tasks) suggest this generation was tuned against real developer workflows rather than generic benchmarks.
- Availability spans nearly every surface Copilot ships on, including the CLI and the cloud coding agent, and it's accessible on the free tier via auto-selection — meaning the vision capability isn't gated behind a paid plan.
Reel Script
Hook (16s)
GitHub just gave Copilot a coding model that can look at a screenshot of your broken UI and actually reason about it — while also being 73% cheaper and 25% faster than the one it replaces.
Core Concept (70s)
Most coding models only ever see text — your source code, your terminal output, your prompt. But a huge amount of real debugging and UI work starts with something visual: a screenshot of a layout that's broken, a hand-sketched wireframe, a diagram of how a system should flow. MAI-Code-1.1-Flash adds native vision support, meaning the model can take an image as direct input in the same conversation as your code, instead of you having to describe the screenshot in words first and lose all the detail in translation. Pair that with the efficiency numbers Microsoft reports: 25% fewer tokens consumed per task and 25% faster token streaming compared to the previous version. Fewer tokens means the model is doing the same job with less wasted reasoning, and faster streaming means you see the response appear on screen quicker — those are two different levers, and this model pulls both at once.
Hands-On (65s)
Here are the actual before-and-after numbers versus the prior MAI-Code-1-Flash: list price down 73%, token streaming speed up 25%, tokens consumed per task down 25%, CLI-task performance up 22%, and .NET-task performance up 15%. Stack the price cut and the token-efficiency gain together and the real cost-per-task drop is bigger than the 73% price cut alone implies — you're paying a lower per-token rate for a task that also needs fewer tokens to finish. And it's not locked behind an enterprise tier: Free and Student users get it automatically through Copilot's auto-model-selection, while Pro, Business, and Enterprise users can select it directly, across VS Code, Visual Studio, JetBrains, Eclipse, Xcode, GitHub Mobile, the Copilot CLI, and GitHub's cloud coding agent.
Takeaway (22s)
For everyday coding tasks — especially CLI work and UI debugging where a screenshot actually helps — this is a genuine efficiency upgrade, not a marketing refresh, since the cost and speed gains compound rather than trade off against each other. If you're on Copilot, switch to it manually today rather than waiting for auto-selection to catch up.