GitHub Copilot Models 2026: GPT, Claude, Gemini Guide

What “GitHub Copilot model” means in 2026

GitHub Copilot stopped being a single model years ago. In 2026 it is a multi-model platform. The model picker in VS Code, JetBrains, the Copilot CLI, and mobile shows options from OpenAI, Anthropic, and Google, plus local models and open-weight models served through GitHub Models. The choice matters more than ever because each model has different strengths, context windows, and pricing tiers.

The official GitHub Copilot documentation describes the model selector as a way to choose which AI model powers chat, code completion, and agentic tasks. The docs also note that some models are marked as “premium” and consume premium requests, which are separate from standard requests. That distinction drives cost and availability.

This guide maps each model to the task it actually wins at. It also shows how to lock your choice in settings so you stop trusting the auto picker.

Every model available in GitHub Copilot in 2026

GitHub Copilot in 2026 offers several model families. The exact list changes through the year, but the core set includes:

• OpenAI GPT-5.x family — includes GPT-5 Pro, GPT-5.2 Pro, GPT-5.4 Pro, GPT-5.5 Pro, and the faster Flash variants. These are general-purpose models strong at code generation and chat.
• Anthropic Claude 4.x family — includes Claude Opus 4, Claude Opus 4.1, Claude Opus 4.6, Claude Opus 4.7, and the “-fast” versions. Opus-class models are known for complex reasoning and refactoring.
• Google Gemini 3 Pro and Flash — Gemini 3 Pro has a large context window, good for long files and repository-wide analysis. Flash is faster and cheaper, suited for quick tasks.
• Open-weight models via GitHub Models — Qwen, Llama, DeepSeek, and others. These are not always available in every editor, but they appear in the model picker when enabled.

Model availability depends on your plan. The GitHub Copilot pricing page lists Pro, Business, and Enterprise tiers, and the docs note that premium models require premium requests. Some models are only available on Business or Enterprise.

How Copilot selects a model

Copilot has two modes: auto and manual. In auto mode, the system picks a model based on the task. The docs describe this as a heuristic that tries to match the model to the complexity of the request. For simple completions it might pick a fast model; for large refactors it might pick a stronger one.

Manual override lets you force a specific model. The precedence order, from lowest to highest, is:

• Repository settings (a .github/copilot.yml or similar file)
• Editor settings (settings.json in VS Code, or the JetBrains settings)
• CLI flags in the Copilot CLI

In practice, the editor setting wins over the repo file, and the CLI flag wins over everything. If you want a consistent experience across a team, set the model in the repo file. If you want to override for yourself, change your editor setting.

Best model per coding task

No single model wins every task. Here is a practical breakdown based on the model characteristics and the pricing snapshot.

Code generation

For generating new code from a description, Claude Sonnet-class models and GPT-5.x are both strong. The pricing snapshot shows Claude Opus 4.1 at $15/M input and $75/M output, while GPT-5 Pro is $15/M input and $120/M output. GPT-5.5 Pro is $30/M input and $180/M output. The higher price of GPT-5.5 Pro suggests it is the flagship, but for everyday generation you might not need it. Claude Opus 4.1 is cheaper per output token and often produces clean, well-structured code.

If you want speed, the “-fast” variants like Claude Opus 4.6-fast ($30/M in, $150/M out) or GPT-5 Flash (not in the snapshot, but the Flash family is known for low latency) are good for quick snippets.

Refactoring

Refactoring is where Opus-class models shine. Claude Opus 4, 4.1, 4.6, and 4.7 are designed for complex reasoning. The pricing snapshot shows Claude Opus 4.7-fast at $30/M in and $150/M out, which is competitive with GPT-5.5 Pro. For large-scale refactors that touch many files, Opus models tend to keep the overall architecture in mind better than smaller models.

GPT-5.2 Pro ($21/M in, $168/M out) is also a strong refactoring choice, but the higher output price means you pay more for the same number of tokens.

Debugging

Debugging often requires reading long logs and many files. Gemini 3 Pro has a large context window, which helps when you need to feed an entire stack trace plus the relevant source files. The pricing snapshot does not list Gemini, but the model is available in Copilot. If you are debugging a memory leak or a race condition, Gemini’s long context lets you paste the whole file without truncation.

Claude Opus 4.1 is also good at reasoning about error messages, but its context window is smaller than Gemini’s.

Tests and documentation

For writing tests and documentation, you do not need the most expensive model. Fast models like GPT-5 Flash or Gemini Flash are cheaper and quicker. The pricing snapshot does not list Flash models, but the pattern is clear: use the cheaper, faster models for repetitive tasks. Writing a unit test for a simple function does not require Opus-level reasoning.

If you are writing documentation for a large API, you might want a model with a large context to see all the functions. Gemini 3 Pro is a good fit there.

How to change the model in VS Code, JetBrains, and Copilot CLI

Changing the model is straightforward, but the exact steps differ by editor.

VS Code

In VS Code, you can change the model via the model picker in the chat panel, or by editing settings.json. The docs mention settings like chat.agent.anthropic.model and chat.agent.openai.model. For example, to force Claude Opus 4.1 for agent tasks, you would add:

"chat.agent.anthropic.model": "claude-opus-4.1"

You can also use the /models slash command in the chat input to see and switch models quickly.

JetBrains

JetBrains IDEs have a similar model picker in the AI Assistant panel. The settings are under Settings > Tools > GitHub Copilot. You can select the model from a dropdown. The docs note that JetBrains supports the same model families as VS Code, but the setting names differ.

Copilot CLI

The Copilot CLI uses a config.toml file. You can set the model per command or globally. For example, to use GPT-5 Pro for a specific request:

copilot --model gpt-5-pro "explain this code"

To set a default model, edit ~/.config/github-copilot/config.toml and add a model = "claude-opus-4.1" line under the appropriate section.

Pricing, premium requests, and rate limits 2026

GitHub Copilot pricing in 2026 still has three tiers: Pro, Business, and Enterprise. The official pricing page lists the monthly cost per user, but the exact numbers are not in the snapshot. What is clear is that premium models consume premium requests, which are a separate quota from standard requests.

The pricing snapshot gives per-token costs for the underlying models, which helps you understand why premium requests are limited. For example, Claude Opus 4.7-fast costs $30/M input and $150/M output. GPT-5.5 Pro costs $30/M input and $180/M output. These are significantly higher than older models like GPT-4 ($30/M in, $60/M out) or o1 ($15/M in, $60/M out).

If you are on a Pro plan, you get a limited number of premium requests per month. Business and Enterprise plans have higher quotas. The docs recommend using standard models for routine tasks and saving premium requests for complex ones.

Rate limits also vary by plan. The docs describe different limits for chat, completions, and agentic features. If you hit a rate limit, you can wait or switch to a different model.

What changed in 2026: new models and deprecations

2026 has been a busy year for model releases. The pricing snapshot shows several new models: GPT-5.4 Pro, GPT-5.5 Pro, Claude Opus 4.6, Claude Opus 4.7, and the “-fast” variants. Older models like GPT-4 and o1 are still listed, but they are likely on a sunset schedule.

GitHub has a model versioning policy. The docs describe a sunset schedule where older models are deprecated and eventually removed. For example, GPT-4 is still available but may be removed later in 2026. The changelog notes that deprecated models stop receiving updates and may be removed after a notice period.

If you have pinned a specific model in your settings, check the deprecation list. The docs recommend updating to a newer model before the sunset date to avoid unexpected failures.

FAQ

Which GitHub Copilot model is best?

There is no single best model. For code generation, Claude Opus 4.1 and GPT-5 Pro are both strong. For refactoring, Claude Opus 4.7-fast is a good choice. For debugging with long logs, Gemini 3 Pro’s large context helps. For tests and docs, use a fast model like GPT-5 Flash or Gemini Flash to save premium requests.

Can I use local models?

Yes, GitHub Copilot supports local models in some editors. The docs describe a feature that lets you connect to a local model server, such as Ollama or LM Studio. This is useful for privacy or offline work, but local models are generally less capable than the hosted ones. The model picker will show local models if you have configured them.

How do I change the model in VS Code?

Use the model picker in the chat panel, or edit settings.json with a setting like chat.agent.anthropic.model. You can also use the /models slash command to switch quickly.

How this review was researched

This review is based on the official GitHub Copilot documentation, the GitHub changelog, the GitHub Copilot pricing page, and live pricing data from the model API providers. No hands-on testing was performed. The model names and prices are taken directly from the pricing snapshot. For more context, see our GitHub Copilot guide, GitHub Copilot news, GitHub AI features, plan comparison, and pricing breakdown.