Foreword¶

In late July 2026, GitHub Copilot released two model integration announcements in quick succession: on July 24, Claude Opus 5 is now available in GitHub Copilot, and on July 28, Grok 4.5 follows up with integration. The flagship inference models from two model vendors entered Copilot’s model picker within nearly the same week, covering mainstream entry points such as VS Code, Copilot CLI, and JetBrains.

For developers who write code daily, this means Copilot is no longer just a completion tool with “one built-in default model”, but is gradually evolving into an Agent platform where you can switch models in your IDE based on tasks. This article organizes the key facts and usage points of this update based on GitHub’s official Changelog and documentation.

July Changelog Blitz: Coexistence of Multiple Models at the IDE Layer¶

If we extend the timeline, Copilot’s multi-model strategy has been quite intensive in July. In addition to Opus 5 and Grok 4.5, on July 9, GitHub also announced that OpenAI’s GPT-5.6 Sol, Terra and Luna are now available in GitHub Copilot. Going further back, Claude Sonnet, Gemini, Kimi K2.7 Code and others have also appeared on the list of models supported by Copilot.

The two announcements from July 24–28 attracted attention because the flagship models from Anthropic and xAI entered the same set of IDE entry points almost simultaneously, and GitHub emphasized their different focuses on Agent workflows during internal testing—Opus 5 favors long-chain reasoning and verification, while Grok 4.5 favors fast parallel tool calling. IDEs are becoming the main battlefield for model vendors to compete for Agent scenarios, and Copilot is playing the role of an “aggregation and distribution layer”.

Claude Opus 5: Complex, Long-Running Agent Tasks¶

According to the July 24 Changelog, Claude Opus 5 is Anthropic’s latest Opus series model, targeted at complex coding tasks that require careful reasoning, effective tool usage, and reliable execution across multiple steps.

GitHub’s conclusions from early testing can be summarized in three points:
1. Excellent Agent workflow performance: Performs well in scenarios such as autonomously modifying code, regression verification, and coordinating multiple tools.
2. More precise modifications: Excels at targeted changes and verifies results after completion, reducing unnecessary execution overhead in complex tasks.
3. Tighter security policies: The model has enhanced protection against high-risk network-related content, and some security-related requests may be blocked; if blocked, the official recommendation is to supplement benign context or switch to another model.

Available plans: Copilot Pro+, Max, Business, Enterprise (excluding the Pro tier).
Pricing: Usage-based billing, settled at the model provider’s API list price. For details, see the Copilot Models and Request Pricing Guide.

Grok 4.5: Large Context Window and Parallel Tool Scheduling¶

The Grok 4.5 announcement was released on July 28. xAI positions it as an inference model for fast Agent coding and complex multi-step workflows, and several hard metrics are worth noting:
- Context window: Up to 500,000 tokens
- Input types: Supports text and images
- Inference tiers: Three reasoning effort options available: low, medium, high

During GitHub’s internal testing, Grok 4.5 performed well in the VS Code terminal scenario and Copilot CLI, especially in parallel scheduling tools and direct action execution, making it suitable for exploratory tasks, complex troubleshooting, and time-sensitive development workflows.

Available plans: Copilot Pro, Pro+, Max, Business, Enterprise (covers the Pro tier, which is more than Opus 5).
Pricing: Also billed based on the provider’s list price plus usage.

Model Picker: Single Entry Point, Consistent Across All Platforms¶

Both models can be switched via Copilot’s model picker, and the official listed entry points highly overlap:

Note: The rollout is gradual. If you cannot see the new models temporarily, please try again later. Some models have minimum requirements for IDE/plugin versions. For example, the official documentation states that Claude Opus 5 requires Copilot extension v1.128.0 or higher on VS Code.

For JetBrains users, there was a supporting update on July 27: GitHub Copilot for JetBrains adds improved OpenTelemetry configuration and model management. It mentions that you can enable or disable all built-in Copilot models in the plugin settings, and configure default values for maxInputToken and maxOutputToken for BYOK/custom endpoints—this aligns with the enterprise-side management requirement of “opening models according to policies”.

Enterprise Administrators: Policy Switches Are Off by Default¶

For Copilot Business and Enterprise organizations, both new models are not “automatically available to all users”:
- Claude Opus 5: Administrators need to enable the corresponding policy in Copilot settings.
- Grok 4.5: Also requires administrator activation and is off by default.

This is consistent with GitHub’s overall approach of “configuring AI model access permissions per organization” in its documentation: enterprises can first evaluate costs, compliance, and scenarios before deciding whether to open specific vendor models to their teams. Individual Pro/Pro+/Max subscribers can try them directly in the model picker (still subject to rollout progress).

How Developers Can Use It: Choose Models Based on Tasks¶

Combined with the official descriptions, we can provide a simple selection reference (not an official recommendation, for reference only in practice):

Scenarios more suitable for Claude Opus 5
- Large-scale refactoring across multiple files that requires multi-round reasoning
- Agent tasks that require running verification and regression checks after modifying code
- Long conversations with high requirements for output stability and step completeness

Scenarios more suitable for Grok 4.5
- Rapid exploration and parallel tool calling in Copilot CLI/terminal
- Reading code bases or long documents with an ultra-large context window (up to 500K tokens)
- Requiring multimodal inputs such as uploaded screenshots to assist in understanding UI/error reports
- Wanting to manually adjust the balance between speed and reasoning depth

In VS Code, you can usually switch models in the model picker of the Copilot Chat panel; Copilot CLI also provides a model picker. For cloud agent scenarios, if the entry point supports model selection, you can specify the model before initiating the task; if not, it will fall back to Auto automatic model selection. For details, see Changing the AI model for Copilot cloud agent.

If you are using Copilot Business/Enterprise and the new models do not appear in the model list, first contact your administrator to confirm whether the policy has been enabled, and check if your local IDE/Copilot plugin is up to date.

Summary¶

GitHub’s consecutive announcements of Claude Opus 5 and Grok 4.5 joining Copilot from July 24–28 are a clear signal of multi-model competition for IDE-layer Agents: developers do not need to leave JetBrains or VS Code, and can switch models from Anthropic, xAI, OpenAI, Google and other vendors within the same Copilot workflow.

For teams, the model picker brings flexibility but also governance issues—who can use which model, and how much it costs, all need to be managed through Copilot policies and usage billing. For individual developers, a more practical approach is to try one model for long-chain Agent tasks and another for CLI rapid exploration, and make choices based on task performance rather than brand preference.

Official documentation and Changelogs will continue to update the model list and minimum version requirements. It is recommended to bookmark Supported AI models in GitHub Copilot to follow subsequent changes.