AI coding cost calculator: GitHub Copilot vs Cursor vs Devin Desktop (Windsurf)
Price one AI coding session on GitHub Copilot, Cursor and Devin Desktop (formerly Windsurf) side by side. Pick a model, enter the tokens one model call sends and gets back, the calls in a session and the sessions in a month, and choose a plan on each product. You get the list price of that work, what each plan already covers and what you would pay on top.
The per-token rates are each product's own published prices (USD per 1 million tokens), read on November 14, 2026. For a model that more than one product lists, the rates were the same on every product that day. What differs is what the plan includes. Everything is calculated in your browser. Nothing you enter is sent or stored.
Start from an example (our assumed token counts, not measurements):
Everything sent to the model: your prompt, open files, instructions, tool definitions and the conversation so far.
An agent resends most of the same context on every call, so much of it can come from the cache. A new chat question usually has none.
1 for a chat question. An agent makes one call per step, often dozens.
Loading the rates...
The same session on every model
List price of one session, cheapest first. A dash means the model is not in that product's list we use. On Cursor, its own models (marked "pool") cost nothing extra while your weekly pool lasts; on Devin Desktop, every model costs nothing extra while your quota lasts. Your selected model is highlighted.
| Model | GitHub Copilot | Cursor | Devin Desktop | Same price? |
|---|
How the estimate works
- Formula: for one call, (uncached input × input rate + cached input × cache-read rate + output × output rate) ÷ 1,000,000 gives dollars. A session is that times the calls, and a month is that times the sessions.
- Cache writes are left out. Several models also charge for writing to the cache, and none of the products says how many tokens a session writes, so estimates with caching are a floor.
- GitHub Copilot: each plan includes a monthly allowance of AI credits (1 credit = $0.01) that works on every model. Past it you pay $0.01 a credit, if you have set a budget for additional usage. Business and Enterprise credits are per user and pooled across the organization. Copilot's 10% discount for auto model selection is not applied here. The Copilot rates are the same file our Copilot credits calculator uses.
- Cursor: Cursor's own models (Grok and Composer) draw from a weekly pool whose size Cursor does not publish, only as multiples of Plus. Past the pool they are paid from prepaid credit at these rates. Third-party models are never in the pool: on Premium and above they are paid from prepaid credit from the first request, and Free and Plus can't use them. Teams seats add the Cursor Token Rate ($0.25 per million input, cached and output tokens) on third-party models unless their cache-read price is below $0.10 per million in every pricing tier.
- Devin Desktop: plans have a daily and weekly quota (Max: weekly only) whose size Devin does not publish, and Devin says quota use can differ from these rates. Past the quota, paid plans buy extra usage at these rates; Free waits for the reset. Models offered free for a time don't count against the quota.
- Ranges: because the Cursor pool and the Devin quota have no published size, "you pay beyond the plan" runs from $0 (all of it fits) to the full list price (none of it fits).
- Long context: some models list higher rates when one call's input, cached tokens included, is above a threshold; we price the whole call at those rates. Where a product names a surcharge without all the rates, the calculator shows "not estimated" instead of guessing.
- Dated prices: introductory and promotional prices switch to the announced later price on the date the product gives (by your clock, in UTC). If no later price is published, the model is left out from that date instead of showing an old price.
- Model lists: Copilot's full list of text models, Cursor's models shown by default, and the models Devin's pricing table recommends. Each product lists more variants (effort levels, fast modes, older models) on its own pages.
- Not included: tax, code completions, Copilot code review, Cursor's 10% uplift for regional data residency, fast and priority variants on Devin, and any rounding the products apply. Your usage page on each product is the record of what you actually used.
Step by step, with the same example session: GitHub Copilot AI credits explained, Cursor pricing explained and Windsurf (Devin Desktop) quotas explained. The products' own pages: GitHub Copilot models and pricing, Cursor models and pricing, Cursor pricing, Devin Desktop models, Devin pricing.
FAQ
Is the same model cheaper on Copilot, Cursor or Devin Desktop?
Per token, no. On November 14, 2026, every model that more than one of the three listed (for example Claude Sonnet 5.5, Claude Opus 5.5, GPT-6 Sol or Gemini 3.8 Flash) had the same input, cache-read and output rates on each of them. The difference is what the plan includes: Copilot plans include AI credits that work on every model, Cursor plans include a weekly pool for Cursor's own models only, and Devin Desktop plans include a quota whose size is not published.
Why does Cursor show the full price for Claude, GPT or Gemini?
Cursor's weekly pool covers only its own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5). Third-party models are paid from prepaid credit at API list price from the first request on Premium and above, and Free and Plus can't use them. Teams seats also pay a Cursor Token Rate of $0.25 per million tokens on most third-party models.
Why is the result a range for Cursor's own models and for Devin Desktop?
Neither product publishes the size of what the plan includes. Cursor describes its pools only as multiples of Plus, and Devin describes its quotas as daily and weekly without a number and says quota use can differ from its per-token rates. So the calculator shows $0 if all of the work fits inside the plan and the full list price if none of it does.
How much does a Copilot plan cover?
Copilot Pro ($10 a month) includes 1,500 AI credits, Pro+ ($39) 7,000 and Max ($100) 20,000, at $0.01 a credit, usable on any model. Business includes 1,900 and Enterprise 3,900 credits per user, pooled. Past the allowance you need a budget for additional usage or a bigger plan.
Are cache writes included?
No. Several models charge for writing to the cache, but none of the three products says how many tokens a session writes, so estimates with caching are a floor.
What happens when a promotional price ends?
The calculator uses your device's date in UTC. On the date a product gives, a model switches to its announced later price. If no later price is published (for example GPT-5.6 Sol on Cursor after November 21, 2026), the model drops out of that product's column instead of showing an old price.
Is anything I enter sent anywhere?
No. The page loads two rate tables from this site and does the arithmetic in your browser. The numbers you enter stay in the page and in the link after # if you copy it.