Copilot Models

GitHub Copilot Model Guide

A compact reference for choosing models in Copilot Chat and agent workflows. Use it to balance model quality, premium request cost, latency, and the type of work you are asking Copilot to do.

Compare Models Use Recommended Flows Return to AI Hub

Reference Table

Available Copilot Models and Cost

Paid-plan cost is shown as GitHub's premium request multiplier. Copilot Free has a limited monthly premium request allowance; included models may not consume premium requests on paid plans.

Model Provider Status Paid Cost Free Plan Best At Use It For
GPT-4.1 OpenAI Retires Jun 1 Included Included Fast everyday coding help, explanations, small changes, and fallback chat. Use only while migrating routine workflows to GPT-5 mini or GPT-5.5.
GPT-4o OpenAI Billing ref Included 1 request General chat, multimodal help, and broad assistant tasks where available. Use as a baseline included model if it appears in your Copilot model picker.
GPT-5 mini OpenAI GA Included Included General coding, writing, quick reasoning, and low-cost daily usage. Default to this for routine prompts before escalating to premium models.
GPT-5.2 OpenAI Retires Jun 1 1x 1 request Deep reasoning, debugging, architectural analysis, and tradeoff decisions. Use only while migrating planning prompts to GPT-5.5 or GPT-5.4.
GPT-5.2-Codex OpenAI Retires Jun 1 1x 1 request Long-horizon coding work, multi-file implementation, and agent mode tasks. Use only while migrating repository-edit workflows to GPT-5.3-Codex.
GPT-5.3-Codex OpenAI GA 1x 1 request Agentic software development, tests, refactors, reviews, and codebase navigation. Use as a strong default for implementation-heavy agent workflows.
GPT-5.4 OpenAI GA 1x 1 request Complex reasoning, debugging, code review, and architecture-level analysis. Use when you need a strong reviewer or planner without jumping to premium-heavy cost.
GPT-5.5 OpenAI GA 7.5x 7.5 requests Frontier reasoning, difficult architecture decisions, and high-stakes code review. Use narrowly for ambiguous planning, security-sensitive review, or GPT-4.1 migration checks.
GPT-5.4 mini OpenAI GA 0.33x 1 request Agentic codebase exploration, search-heavy work, and lightweight implementation. Use to scan a repo, find likely files, draft a plan, or handle medium-risk edits.
GPT-5.4 nano OpenAI GA 0.25x 1 request Low-cost snippets, short explanations, small rewrites, and quick checks. Use for cheap iteration when mistakes are easy to spot and fix.
Claude Haiku 4.5 Anthropic GA 0.33x 1 request Fast answers, lightweight edits, small explanations, and cheap support loops. Use for quick iteration before escalating to Sonnet or Opus.
Claude Sonnet 4.5 Anthropic GA 1x 1 request General coding, agent tasks, reasoning, and practical development support. Use as a balanced implementation model when you want quality at 1x.
Claude Sonnet 4.6 Anthropic GA 1x 1 request General coding, debugging, visual reasoning, and agentic development. Use as the main "build it" model after a harder model has planned the approach.
Claude Opus 4.5 Anthropic GA 3x 3 requests Difficult refactors, codebase reasoning, and complex problem solving. Use for architectural planning when Sonnet is missing context or nuance.
Claude Opus 4.6 Anthropic GA 3x 3 requests Deep reasoning, hard debugging, and complex tradeoff analysis. Use for planning and review on complicated work when 3x cost is acceptable.
Claude Opus 4.6 fast mode Anthropic Preview 30x 30 requests Low-latency Opus-class reasoning when response speed matters more than budget. Reserve for rare interactive sessions where premium request burn is acceptable.
Claude Opus 4.7 Anthropic GA 15x 15 requests Anthropic's strongest Copilot model for advanced coding and reasoning. Use for the first planning pass on large, ambiguous, or risky engineering tasks.
Gemini 2.5 Pro Google GA 1x 1 request Complex code generation, debugging, research, and broad technical analysis. Use as a strong alternative reviewer or researcher at standard cost.
Gemini 3 Flash Google Preview 0.33x 1 request Fast simple tasks, repetitive edits, summaries, and low-cost iteration. Use for quick work when latency and cost matter more than depth.
Gemini 3.1 Pro Google Preview 1x 1 request Long-context reasoning, edit-test loops, and technical analysis. Use for reading large contexts, comparing alternatives, and research-heavy prompts.
Grok Code Fast 1 xAI Deprecated 0.25x 1 request Fast coding, code generation, and debugging across common languages. Avoid for new workflows; move speed-focused prompts to GPT-5 mini or Claude Haiku 4.5.
Raptor mini Fine-tuned GPT-5 mini Preview Included Included Inline suggestions, explanations, and everyday developer assistance. Use for included-cost support when available in your environment.
Goldeneye Fine-tuned GPT-5.1-Codex Preview Free only Included Evaluation-oriented reasoning and problem solving. Use for comparison and testing when GitHub exposes it in your picker.

Current caveat: GitHub may auto-select a model for some requests, especially if the selected model is overloaded. Models also vary by Copilot plan, client, policy settings, and preview access.

Task Guide

Quick Picks by Task Type

Start with the cheapest model that can do the job, then escalate only when the task is ambiguous or risky.

Small code edits

Start: GPT-5 mini, Claude Haiku 4.5, or GPT-5.4 nano.

Escalate: Claude Sonnet 4.6 or GPT-5.3-Codex if the edit touches several files.

Feature implementation

Start: Claude Sonnet 4.6 or GPT-5.3-Codex.

Escalate: Claude Opus 4.7 or GPT-5.4 for the plan if the feature is unclear.

Debugging production issues

Start: GPT-5.4 or Gemini 2.5 Pro.

Escalate: Claude Opus 4.7 for difficult root-cause analysis.

Large refactors

Start: Claude Opus 4.7 or GPT-5.4 for scope and risk analysis.

Implement: Claude Sonnet 4.6 or GPT-5.3-Codex in smaller verified chunks.

Code review

Start: GPT-5.4, Gemini 2.5 Pro, or Claude Sonnet 4.6.

Escalate: Claude Opus 4.7 for high-risk logic, security-sensitive areas, or architecture drift.

Docs and explanations

Start: GPT-5 mini, Gemini 3 Flash, or Claude Haiku 4.5.

Escalate: Gemini 3.1 Pro for long context or GPT-5.4 for technical precision.

Visual or UI reasoning

Start: Claude Sonnet 4.6, Gemini 2.5 Pro, or GPT-5.4.

Escalate: Opus when layout, product behavior, and code changes all interact.

Budget-sensitive loops

Start: included models, GPT-5.4 nano, Claude Haiku 4.5, or Gemini 3 Flash.

Escalate: Pay 1x or higher only after the problem is narrowed.

Recommended Flows

How to Chain Models

Use different models for different phases instead of trying to make one model do every job.

Complex feature flow

  1. Plan with Claude Opus 4.7 or GPT-5.4: define scope, risks, file map, and test strategy.
  2. Implement with Claude Sonnet 4.6 or GPT-5.3-Codex: make the changes in small commits or agent passes.
  3. Review with GPT-5.4 or Gemini 2.5 Pro: look for missed edge cases, broken assumptions, and missing tests.
  4. Polish with GPT-5 mini or Claude Haiku 4.5: tighten copy, comments, and small follow-up edits.

Bug triage flow

  1. Summarize with GPT-5 mini: collect symptoms, logs, repro steps, and suspected areas.
  2. Diagnose with GPT-5.4, Gemini 2.5 Pro, or Claude Opus 4.7: reason through likely causes.
  3. Patch with Claude Sonnet 4.6 or GPT-5.3-Codex: edit the smallest safe surface area.
  4. Verify with GPT-5.4: inspect the diff, tests, and regression risk before merging.

Large refactor flow

  1. Inventory with GPT-5.4 mini or Gemini 3.1 Pro: find call sites, dependencies, and duplicate patterns.
  2. Design with Claude Opus 4.7: choose the migration path and define boundaries.
  3. Execute with GPT-5.3-Codex or Claude Sonnet 4.6: refactor one subsystem at a time.
  4. Audit with GPT-5.4 or Claude Opus 4.7: use a final high-reasoning pass only on the risky diff and tests.

Low-cost daily flow

  1. Ask with included models: use GPT-5 mini or Raptor mini for most small prompts.
  2. Explore with 0.25x to 0.33x models: use GPT-5.4 nano, Gemini 3 Flash, or Haiku.
  3. Build with 1x models: use Sonnet 4.6, GPT-5.3-Codex, GPT-5.4, or Gemini 2.5 Pro when files change.
  4. Escalate only for ambiguity: reserve GPT-5.5 and Opus 4.7 for decisions that affect architecture or reliability.

Cost Notes

How to Think About Cost

Premium request multipliers are not dollar prices. They are how quickly a request consumes your included or purchased premium request allowance.

Use included models first

GPT-5 mini and Raptor mini are the safest starting point for routine prompts because paid-plan usage is included according to GitHub's multiplier table. GPT-4.1 should be treated as a migration-only option.

Treat 1x as the default work tier

Sonnet, Codex, GPT-5.4, and Gemini Pro models are usually where implementation, review, and deeper analysis should happen.

Reserve premium-heavy models

GPT-5.5, Claude Opus 4.7, and Opus fast mode should answer narrow, high-value questions rather than carrying every iteration of a long coding session.

Sources

Official References

The model list changes often. Re-check these sources before making budget or policy decisions.

Supported models GitHub's current supported model list and premium request multiplier table.
Model comparison GitHub's capability comparison and task-fit guidance across Copilot models.
Premium requests How premium requests work today and how usage-based billing changes on June 1, 2026.
Models and pricing GitHub's June 1 AI-credit pricing table for Copilot model usage.
Model hosting Where OpenAI, Anthropic, Google, xAI, and fine-tuned GitHub models are hosted.
Model deprecation notice GitHub's May 1, 2026 notice for GPT-5.2 and GPT-5.2-Codex deprecation on June 1.
GPT-4.1 deprecation notice GitHub's May 7, 2026 notice that GPT-4.1 leaves Copilot on June 1.
Grok deprecation notice GitHub's May 15, 2026 notice that Grok Code Fast 1 is deprecated across Copilot.