GitHub Copilot July Update Adds Vision and AI Credit Limits

What arrived all at once?

GitHub published several Copilot-related Changelog posts on July 1. First, Copilot Vision moved to general availability. Users can attach images and PDFs to Copilot Chat and have them interpreted alongside code.

On the same day, browser tools for GitHub Copilot in VS Code were also announced as generally available. This is a workflow where an agent opens a real browser, explores a web app, and brings the results back into the conversation. In other words, the screen and browser state can now become part of the working context.

Why cost control matters more now

Another key point is that AI credit session limits can now be set in Copilot CLI and the Copilot SDK. According to GitHub, this lets users set an upper limit on how many AI credits an agent can use in a single session, helping prevent excessive usage.

This is not just a billing option. Coding agents run commands, revise their work when something fails, read multiple files, and repeat tasks. In that kind of workflow, it becomes harder to predict how long a single session will run. That is why the ability to set a budget line per task has become practically important.

Model selection is moving toward automation

GitHub also said enterprise administrators can set Copilot’s default model to auto model selection. Instead of manually choosing a model for each new conversation, the workflow shifts toward automatic routing based on the task.

It is also worth noting that Kimi K2.7 Code became generally available as a model option in GitHub Copilot. GitHub describes it as the first open-weight model added to the Copilot model picker. In other words, Copilot is becoming less like a fixed single-model tool and more like a platform that chooses among multiple models based on the task and policy.

What developers should check

It would be risky to read this update as meaning that “all work has automatically become easier.” Image and PDF attachments, browser operation, and automatic model selection are convenient, but they also mean more project code, documents, and screen information can enter Copilot’s context.

Individual developers should check feature support by environment and credit consumption. Team administrators should review allowed models, browser access, attachment policies, and session limits. In short, this GitHub Copilot update is a signal that AI coding tools are entering a stage where they become smarter and require operational management at the same time.