Xcode 26 introduces Intelligence Mode, which lets developers use AI to assist with coding tasks. It includes a new Agent Mode that allows the AI to interact with your codebase and provide context‑aware suggestions.
In Xcode 26, Apple introduced Intelligence Mode — an AI‑powered assistant that provides context‑aware code suggestions to help you write code faster and with fewer errors. As of this post, Intelligence Mode is available in the Xcode 26 beta. It expands on traditional code completion (the system in Xcode 16) by offering richer, project‑aware recommendations rather than only inline completions.
A key part of Intelligence Mode is Agent Mode, which lets the AI interact directly with your codebase (for example, creating, editing, or removing files) under developer control. Agent Mode is designed to accelerate common tasks and surface relevant, project‑specific changes while preserving developer oversight and review.
The default model in Intelligence Mode is OpenAI ChatGPT. Apple hasn’t disclosed the exact variant (it may match the model used by Apple Intelligence when Siri hands off a query), and there’s a vague daily usage limit. For development, this default can be problematic: ChatGPT‑style models often produce plausible‑looking but incorrect code, can miss project‑specific constraints, and aren’t guaranteed to be optimized for reliable refactors or build‑safe edits. GPT‑5 is launching and promises stronger programming capabilities, but Apple has not said Xcode uses GPT‑5 (or any specific release), so for critical development work you should consider configuring a known, code‑focused model you can verify and control (local or hosted).
This post shows how to add custom models to Xcode 26 Intelligence and expand the assistant’s capabilities. You may prefer a specific model — for example Claude Sonnet 3.7, the state‑of‑the‑art Claude Sonnet 4, or Claude Opus 4 — and this guide explains how to configure Xcode to use them (both local and hosted) so you can verify and control model behavior in Agent Mode.
Anthropic’s Claude family (Sonnet, etc.) is widely regarded as one of the strongest options for code generation and refactoring. These models tend to produce clearer, more context‑aware edits, handle long context windows well, and follow instructions conservatively, which reduces brittle or build‑breaking changes. They also perform well at tool‑calling patterns, though that is less relevant today because Xcode 26’s Agent Mode does not yet expose tool‑calling capabilities.
Even though Anthropic’s Claude models are among the best in the industry, they are also some of the most expensive options. However, the cost reflects the quality — you get what you pay for. These models deliver exceptional performance, especially for complex, context‑heavy tasks, making them a worthwhile investment for critical development projects where accuracy and reliability are paramount.
To start using the Anthropic model in Xcode, follow these steps:
Create an Anthropic account Visit the Anthropic Console and register for an account if you don’t already have one.
Generate an API key After logging in, navigate to the API Keys section in the dashboard and create a new API key. Make sure to copy and securely store the key, as you’ll need it to configure Xcode.
Configure Xcode to use the API key Open Xcode 26, go to Preferences > Intelligence Mode, and select Add a Model Provider. Choose Internet Hosted. Enter your Anthropic API key in the provided field.
Select the Desired Model If successful, you should see a list of available models.
Test the model
Once configured, you can select the Anthropic model in Xcode 26 and start using it in Agent Mode. Open the Coding Assistant panel or press ⌘+0 to begin.
OpenAI’s GPT‑5 is another strong option for code generation and refactoring. It promises improved programming capabilities, but it’s not yet confirmed. If you want to use OpenAI’s GPT‑5 in Xcode, you can follow similar steps as with Anthropic:
Create an OpenAI Platform account Visit the OpenAI Console and register for an account if you don’t already have one.
Generate an API key After logging in, navigate to the API Keys section in the dashboard and create a new API key. Make sure to copy and securely store the key, as you’ll need it to configure Xcode.
Configure Xcode to use the API key Open Xcode 26, go to Preferences > Intelligence Mode, and select Add a Model Provider. Choose Internet Hosted. Enter your OpenAI API key in the provided field.
Select the Desired Model If successful, you should see a list of available models.
NOTE: I needed to manually add the allowed model in my OpenAI account; otherwise it did not show up in the list.
This is accessible in Limits → Allowed Models → Edit.
Test the model
Once configured, you can select the OpenAI model in Xcode 26 and start using it in Agent Mode. Open the Coding Assistant panel or press ⌘+0 to begin.
If you have a local LLM running, such as Ollama or LM Studio, you can also use it in Xcode 26. Here’s how to set it up:
Configure Xcode Open Xcode 26, go to Preferences > Intelligence Mode, and select Add a Model Provider. Choose Locally Hosted. Enter the port and description.
Select the Desired Model If successful, you should see a list of available models.
Test the modelOnce configured, you can select the local model in Xcode 26 and start using it in Agent Mode. Open the Coding Assistant panel or press ⌘+0 to begin.
Another option is to use an AI middleware aggregator, which allows you to configure a single API key and access multiple models (such as Claude, OpenAI GPT, and others) seamlessly.
One popular choice is OpenRouter.io, which supports models from various AI providers. These include Anthropic for its Claude models, OpenAI for its GPT models, Google with Gemini, DeepSeek, Qwen3, Kimi K2, Meta for its LLaMA models, and even GPT-OSS. OpenRouter provides detailed documentation on integrating with Xcode 26, which you can find here. For a complete list of supported models, visit OpenRouter Models.
I personally use OpenRouter to manage my AI models. It provides a unified interface for accessing various AI capabilities and makes switching between different models seamless. This flexibility is especially valuable when working on diverse projects that require specific AI models for optimal performance. Additionally, the pricing is highly competitive compared to using APIs directly from individual AI providers.
The new Xcode 26 Intelligence feature is an exciting addition to the Xcode ecosystem. I’m genuinely eager to incorporate it into my daily development workflow. However, it still falls short when compared to other AI‑powered code editors like Cursor, VS Code with Copilot, or even Chinese code editors like Trae. In practice, I often find myself relying on these other editors more frequently than Xcode 26, even though Xcode is considered the best IDE for iOS development.
Adding features such as tool‑calling capabilities, customizable system prompts, support for the Model Context Protocol (MCP), and other enhancements would significantly improve Xcode 26. These additions would make it more competitive with other AI‑powered code editors and elevate its utility for developers.
The new Xcode 26 Intelligence feature, particularly Agent Mode, is a powerful tool for developers. By configuring custom models like Anthropic’s Claude or OpenAI’s GPT‑5, you can enhance the AI’s capabilities and ensure it aligns with your project needs. Whether you choose a hosted solution or a local LLM, the flexibility to customize the AI assistant in Xcode 26 opens up new possibilities for efficient and effective coding.
As Xcode 26 continues to evolve, I look forward to seeing how Apple enhances Intelligence Mode and Agent Mode, making them even more powerful and user‑friendly. For now, the ability to integrate custom models is a significant step forward, allowing developers to tailor the AI assistant to their specific requirements and preferences.
