Best AI coding assistant

Coding assistants are wrappers around models, so we score one by running the flagship model the tool ships with through our coding suite. That is only possible when the tool names the model it uses, which is why some of the tools on this page carry a score and some do not. The ones without a score are listed underneath, unranked, with everything we do hold on them.

How this is ranked

Our tested score on the 18-task coding suite — writing new code, finding real bugs, explaining unfamiliar code, marked on correctness first.

2 of the 6 coding tools we track carry that score, so 2 are ranked here. The other 4 are listed below the ranking, unranked, with everything we do hold on them. How the testing works →

GitHub/Microsoft · 18 tasks · 10 Aug 2026 · receipts ↓

It suggests lines of code and functions to help you write software.

Its strongest task against the field — Typed debounce, 9/10 · judged by grok-4.5

Correct typed debounce with cancel; preserves args via Parameters/ThisParameterType; no any; accurate one-sentence explanation. Minor polish only.

Cheapest verified plan
$10/mo (pro) · free version confirmed 17 Aug 2026
What users rate it
No app-store listing we can score
Public conversation (90 days)
Negative 13 of the 13 posts that took a side (108 scanned)
Trains on what you type
GitHub uses your inputs to train its AI models by default, but you can turn this off in your account settings. docs.github.com

full GitHub Copilot page → · visit github.com

#2

Cursor

85/100

Anysphere · 18 tasks · 10 Aug 2026 · receipts ↓

You use it to write, test, and build software.

Its strongest task against the field — Retry with backoff, 10/10 · judged by gemini-3.1-pro-preview

The response perfectly fulfills all requirements, including the constraints. The code is correct, implements exponential backoff with jitter, handles the 4xx logic accurately, and clearly explains the final failure behavior within the…

1 answer broke a checkable rule in the task and was capped in code, not by a judge’s opinion.

Cheapest verified plan
$20/mo (pro)
What users rate it
No app-store listing we can score
Public conversation (90 days)
Mixed 2 positive, 2 negative of 4 that took a side (37 scanned)
Trains on what you type
They will not use your inputs to train their AI models unless you explicitly agree to it. cursor.com

full Cursor page → · visit cursor.com

Tracked, not ranked

These are real coding tools we follow — prices, ratings, public reception and data policies — but none has a tested score, so none has a position. Placing them would mean inventing one.

Lovable

Lovable

It helps you build websites and applications by describing what you want in chat.

No tested score, so it is not ranked here. We do hold a quotable data-training policy and 164 Hacker News points in 90 days.

Replit

Replit

It helps you build and deploy web and mobile apps without needing to write code.

No tested score, so it is not ranked here. We do hold 4.72★ from 20k reviews, a quotable data-training policy and 19 Hacker News points in 90 days.

v0

Vercel

You use it to build and publish full-stack web applications and websites using text prompts.

No tested score, so it is not ranked here. We do hold a quotable data-training policy.

Windsurf

Cognition

Manage and run local and cloud coding agents to write, test, and ship code.

No tested score, so it is not ranked here. We do hold a quotable data-training policy and 5 Hacker News points in 90 days.

Questions people ask

What is the best AI coding assistant?

GitHub Copilot — it scored 90/100 on our coding suite (18 tasks, marked blind by three judges from three rival AI labs, 10 Aug 2026), ahead of Cursor on 85. We track 4 more coding tools that carry no tested score, so they are listed but not ranked.

How is this ranked?

On our tested score on the 18-task coding suite — writing new code, finding real bugs, explaining unfamiliar code, marked on correctness first. Nothing else is weighted in: not price, not popularity, not how many reviews an app has. Those figures are all on this page because they matter when you choose, but they do not move a position.

Which of these are free?

GitHub Copilot publish a free version on their own pricing pages, confirmed on the dates shown beside each one. GitHub Copilot's cheapest paid plan we could verify is $10 a month; Cursor's cheapest paid plan we could verify is $20 a month.

Why is Lovable not ranked?

Lovable: No tested score, so it is not ranked here. We do hold a quotable data-training policy and 164 Hacker News points in 90 days. Replit: No tested score, so it is not ranked here. We do hold 4.72★ from 20k reviews, a quotable data-training policy and 19 Hacker News points in 90 days. v0: No tested score, so it is not ranked here. We do hold a quotable data-training policy. Windsurf: No tested score, so it is not ranked here. We do hold a quotable data-training policy and 5 Hacker News points in 90 days. We score a tool by running the flagship model it ships with through the suite, which is only possible when the tool names that model. Listing it unranked is more honest than guessing a position for it.

No vendor pays for placement, position or inclusion. The order above is computed from the tested scores, and the code that computes it cannot see which outbound links earn us anything. How we make money →

Side by side: Cursor vs GitHub Copilot vs Windsurf · Same job, ranked by model: Best AI for coding
Every tool head-to-head → · Every tool we track → · Methodology → · Who runs the tests →
Tested and published by Robert Prime · re-ranks by itself when a new run lands · last run 10 Aug 2026