Cursor vs Claude Code vs GitHub Copilot: A Task-First Buying Guide

Revised October 15, 2026: we re-read the pricing sources, added a task-based choice path and corrected the trial so each candidate starts from the same commit. This is document research, not testing we ran. Regional access, contractual rights and enterprise data terms require their own checks.

Choose a workflow before a model#

For an English-speaking individual or small team: investigate Cursor if the editor is your daily workspace; Claude Code for terminal-led repository tasks; GitHub Copilot if keeping GitHub integrations and organisation administration is the priority. These are editorial starting points, not measured code-quality rankings.

Cursor's documentation describes its editor, agents and CLI. Claude Code's overview covers terminal, IDE, desktop and web. GitHub's plan documentation describes IDE, web, CLI and app access. Record the actual model and release in your account; a product name does not fix either.

Cost basis checked October 15#

• Cursor: the pricing page shows Free, Plus at $8/month with Grok models only, and Premium at $30/month. Third-party models are available from Premium and billed separately at API list price; prices exclude applicable taxes. A plan that cannot run your required model is not a like-for-like comparison.
• Claude Code: Claude pricing includes Code in Pro, not Free. Pro is $20 billed monthly or $200 upfront annually, displayed as about $17/month. API billing is separate; see the cost documentation. A vendor average is not your team's quote.
• Copilot: GitHub's plan table lists Pro at USD 10/month and a restricted Free option for eligible individuals. Organisation seats and AI-credit allowances have different cost bases.

These figures are page observations, not currency conversions or worldwide offers. Confirm checkout currency, tax, cycle, region, output rights and code-handling terms. For confidential code, establish training use, retention and administrator permissions before uploading; a local terminal does not mean local-only inference.

Exclusions and alternatives#

Gemini CLI is a terminal alternative if its current quota and model fit. Hosted free usage differs from a software licence. Jules, Codex, Devin Desktop, Trae, Aider and Cline are not ranked below this shortlist: their current plan/deployment costs were not re-verified here. MonkeyCode requires its own deployment evaluation. No candidate is certified here for a policy that prohibits code leaving an approved environment.

A fair trial you can run this week#

This is a proposed protocol. Pick one multi-file change with written pass/fail criteria. Save the starting commit; create a separate worktree or clone from that exact commit for each candidate. Never give the second tool the first tool's implementation.

• Save the same prompt, authorised files, fixtures, acceptance commands and time/spend ceiling. Use equivalent permissions; record model or plan differences.
• Before each run, record date, product/build, model, plan, operating system, provider, starting commit and settings.
• Preserve the original diff, command output, elapsed time, credits/token charges and every human correction. Judge correctness, scope and maintainability before speed.
• Record quota/region problems as blocked, and unsupported environments as excluded with reasons. Retain unsuccessful attempts.
• Compare total human time and cost for an accepted change. Merge through normal review. One task does not establish a universal winner.

Failure cases and next steps#

A tiny task measures little; an oversized task creates an unreviewable diff. Reusing a modified repository contaminates the comparison. API spend and reviewer rework can exceed the subscription. Reject and record unnecessary permission expansion.

Choose only a candidate that meets your actual acceptance criteria, budget and data conditions. If none does, narrow the task or retain your workflow. Historical context: coding annual review, Cursor launch, Copilot coding agent. These do not replace current plans. See editorial standards.