“55% faster.” GitHub’s famous number is real — for one lab task. Randomised trials on real codebases tell a more complicated story. Here’s what the evidence says, and what Cursor and Copilot now actually cost.
Evidence review of randomised trials, field experiments, industry surveys, security research and official pricing, checked 14 Oct 2026. Hyptor Lab’s timed 10-task test is scheduled. No affiliate links. How we test
The short answer
They can make you faster, but less than the marketing implies — one rigorous trial even found experienced developers got slower. Copilot is cheaper and more predictable; Cursor is mid-transition after its SpaceX acquisition. Review the code either way.
Cheapest paid: Copilot Pro $10/mo · Free tier: Copilot’s is clearly defined · Security: treat AI code as untrusted until reviewed · Confidence: Medium — strong research on assistants generally, none comparing these two head to head.
What they claim vs what the evidence shows
| Their claim | Verdict | What the evidence shows |
|---|---|---|
| GitHub: developers are “up to 55% more productive at writing code without sacrifice to quality” [3] | Overstated | The 55% comes from one lab task — writing a JavaScript HTTP server (1h11m vs 2h41m, 95 developers) [8][9]. Quality wasn’t measured. Field experiments found +26% completed tasks [12]. |
| AI assistants make experienced developers faster on real work | Holds, with caveats | METR’s 2025 trial found tasks took 19% longer with AI (mostly Cursor) [10]; its 2026 update shows “some evidence for speedup” but calls the data unreliable [11]. |
| Cursor: “Trusted by over half of the Fortune 500” [17] | Untested | A company claim we couldn’t independently verify; adoption isn’t evidence of productivity. |
Why this matters
Teams buy seats on productivity claims. Code that “looks right” but is insecure or duplicated can cost more time later than it saves now. And both products changed their pricing in 2026, so last year’s comparisons are out of date.
How we built this verdict
We read the original studies behind the productivity claims [8–12], Google’s DORA research [13][14][18], security studies [15][16][19] and both vendors’ pricing and docs [1–7]. We found no independent head-to-head study of Cursor vs Copilot, so we compare what can be verified.
The “55% faster” number comes from one small task
Finding: in GitHub’s 2022 experiment, 95 developers wrote an HTTP server in JavaScript; the Copilot group finished in 1h11m versus 2h41m (95% CI 21–89%) [8][9]. A larger field study across Microsoft, Accenture and a Fortune 100 firm (4,867 developers) found 26.08% more completed tasks with Copilot, with bigger gains for less-experienced developers [12].
In practice: real gains exist, especially on well-defined tasks and for juniors — but “55%” isn’t what most teams should expect.
The surprise: experienced developers got slower
Finding: METR ran a randomised trial with 16 experienced open-source developers on 246 real issues, mostly using Cursor Pro. With AI, tasks took 19% longer — yet developers believed they’d been about 20% faster [10]. METR’s February 2026 update found “some evidence for speedup” (−18% task time, confidence interval −38% to +9%) but calls its own signal unreliable [11].
In practice: how fast you feel isn’t how fast you are. Measure on your own work.
AI-written code needs a security review
Finding: Veracode tested 100+ models and found 45% of code samples failed security tests, failing to defend against cross-site scripting in 86% of relevant samples [16]. An earlier academic study found about 40% of Copilot-generated programs vulnerable [15]. GitClear’s data showed duplicated code blocks rising eight-fold in 2024 [19]. Google’s DORA 2024 report linked rising AI adoption to a 7.2% drop in delivery stability [13], though DORA 2025 linked AI to higher throughput [14].
Where each one genuinely shines
Copilot: the cheapest entry ($10/mo), a defined free tier (2,000 completions and 50 chats a month), and completions that don’t use credits [3][5]. Cursor: an AI-first editor built around multi-file agent work, used in METR’s trials [10]; its new owner promises “more capable models at lower cost” [6].
What they really cost
| GitHub Copilot | Price | Monthly AI credits |
|---|---|---|
| Free | $0 | 2,000 completions + 50 chats |
| Pro | $10/mo | 1,500 (completions unlimited) |
| Pro+ | $39/mo | 7,000 |
| Max | $100/mo | 20,000 |
| Business / Enterprise | $19 / $39 per seat | 1,900 / 3,900, pooled |
Cursor: its pricing page lists Free, Plus $8 (Grok models only), Premium $30, Super $100 and Ultra $200, sold via X Subscription Corp after the SpaceX acquisition [1][6] — but its own docs still list Pro $20, Pro Plus $60 and Ultra $200 [2]. Check the live checkout before buying. Cursor apologised for a confusing pricing change in July 2025 and offered refunds [20].
Which should you use?
Budget or GitHub-centric team: Copilot — clearer pricing and a usable free tier. Want an AI-first editor and agent workflows: trial Cursor, after confirming the current plan terms. Everyone: measure your own speed before and after, and review every AI-written change for security. Note that Copilot may use Free/Pro/Pro+ interactions for training unless you opt out [3].
Questions people ask
Is Cursor better than GitHub Copilot?
No independent head-to-head study exists. Copilot is cheaper with clearer pricing [5]; Cursor is an AI-first editor whose plans are changing after its acquisition [1][2].
Do AI coding assistants really make you faster?
Often, but not always: field studies found +26% completed tasks [12], while a randomised trial with experienced developers found 19% slower [10].
Is AI-generated code secure?
Not by default: 45% of samples failed security tests in Veracode’s 2025 study [16]. Review it like any untrusted code.
The verdict
AI coding assistants deliver real but uneven gains, and the famous 55% figure overstates what most teams will see. The decision: start with Copilot’s free tier, trial Cursor if you want an AI-first editor, measure your actual speed, and review everything.
Change log
15 Oct 2026 — Published as an evidence review. Hyptor Lab timed test scheduled.
Sources
[1] Cursor pricing · [2] Cursor docs, pricing · [3] GitHub Copilot plans & FAQ · [4] GitHub Blog, 27 Apr 2026 · [5] GitHub Docs, Copilot plans · [6] Cursor, “Cursor is now a part of SpaceX”, 14 Aug 2026 · [7] GitHub Copilot features · [8] GitHub Blog, Copilot productivity research · [9] Peng et al., 2023 · [10] METR, 10 Jul 2025 · [11] METR, 24 Feb 2026 · [12] Cui et al., Management Science · [13] DORA 2024 · [14] DORA 2025 · [15] Pearce et al., IEEE S&P 2022 · [16] Veracode, 30 Jul 2025 · [17] Cursor homepage · [18] DORA 2025 report · [19] DevClass on GitClear, 20 Feb 2025 · [20] Cursor, “Clarifying our pricing”, 4 Jul 2025. Accessed 14 Oct 2026.
Right of reply
Cursor and GitHub can respond via our contact page.