AI coding tool pricing compared: every plan, side by side

Every vendor prices agent coding differently and none of them price it in a unit you can compare. Here is all of it converted into the only number that matters: what you pay a month for the way you actually work.

By the Continuum team. We build a workbench that runs Claude Code, Codex, and their peers, so the model rates quoted here are the ones our own cost analytics ship with.

The short version

As of August 2026 the entry prices are: Claude Code from $20/mo with Claude Pro, Codex from $0 because it is included in every ChatGPT plan, Cursor from $20/mo, GitHub Copilot from $10/mo, Gemini Code Assist from $19/user/mo on an annual commitment with no free individual tier since 18 June 2026, and Devin from $20/mo. Heavy tiers converge on $200/mo almost everywhere. The comparison that actually matters is not the sticker price but the billing shape: rolling usage windows, prepaid credit pools, or token quotas with metered overage.

What you need to know
  • Entry prices cluster at $0 to $20. Heavy tiers cluster at $200. The middle is where vendors differ.
  • Codex is the only agent included at $0, because it rides every ChatGPT plan including Free.
  • Gemini Code Assist lost its free individual tier on 18 June 2026. That row changed more than any other this year.
  • Three billing shapes: usage windows, credit pools, token quotas. They fail differently.
  • Two $100 subscriptions often beat one $200 one, because you get two independent windows.

The whole price list

Start here. Every figure was read from the vendor's own pricing page or documentation on 7 August 2026. Where a vendor does not publish a number, the cell says so instead of guessing.

Individual plans. Per user per month unless stated.

VendorFreeEntryMidHeavy
Claude CodeNonePro $20 ($17 annual)Max 5x $100Max 20x $200
CodexYes, $0Go $8, Plus $20No separate published rungPro $200
CursorHobby, limitedPro $20Pro+ price not publishedUltra price not published
GitHub CopilotYes, limitedPro $10Pro+ $39Max $100
Gemini Code AssistDiscontinuedStandard $19 annualStandard $22.80 monthlyEnterprise $45 annual
DevinYes, limitedPro $20Teams seat $40Max $200

Team and business seats, checked 7 August 2026.

VendorTeam pricingNotes
ClaudeStandard seat $20/seat annual, $25 monthly. Premium seat $100 annual, $125 monthlyClaude Code on both seat types
OpenAIBusiness $20/user annual, $25 monthly, 2 users minimumBusiness limits match Plus, per user
CursorTeams $40/userEnterprise is custom
GitHub CopilotBusiness $19/seat, Enterprise $39/seatPer granted seat
Gemini Code AssistStandard $19 annual / $22.80 monthly. Enterprise $45 / $54All subscriptions billed monthly
DevinTeams $80/mo minimum, full seats $40 each, flex seats freeUnlimited members

The three billing shapes

The sticker prices look comparable. The mechanisms do not, and the mechanism decides how the plan fails you.

ShapeWho uses itHow it runs outWhat it costs to keep going
Rolling usage windowsClaude, CodexYou are blocked until the window ages outNothing, but you wait
Credit poolsCursor, GitHub CopilotThe pool emptiesBuy more, or drop to a cheaper model
Token quotasDevin, Gemini Code AssistA daily or weekly budget hits zeroMetered overage at API list prices

What the credit units actually mean

Two vendors publish a dollar-per-credit rate and one does not. GitHub is explicit: 1 GitHub AI Credit equals $0.01, and code completions and next edit suggestions consume none of them, so a Copilot Pro seat is $10 of subscription carrying $15 of credit allowance. OpenAI publishes Codex credit rates per million tokens without publishing a dollar per credit, so treat those numbers as relative weights rather than money.

Codex credit rates per million tokens, as published on OpenAI's Codex pricing page. Credits, not dollars.

ModelInputCached inputOutput
GPT-5.6 Sol12512.5750
GPT-5.6 Terra505300
GPT-5.6 Luna50.530

The ratio is the useful part: Luna costs a twenty-fifth of Sol on input and Terra sits between them. Whatever a credit is worth, model choice moves your consumption by more than an order of magnitude, which is the same lesson every one of these vendors teaches in a different currency.

Three ways vendors meter agent coding: a rolling window that ages back, a credit pool that empties, and a daily token quota that refreshes, each failing in a different wayROLLING WINDOWCREDIT POOLTOKEN QUOTAClaude · CodexCursor · CopilotDevin · Geminiused capacity ages outdrains perpremium turna daily budget, refreshedWHEN IT RUNS OUTWHEN IT RUNS OUTWHEN IT RUNS OUTyou are blockedthe pool emptiesthe budget hits zeronothing, but you waitbuy more, or drop modelAPI list prices applyA window stops you. A pool or a quota keeps going and moves the bill.

What $0 buys, per vendor

Free tiers in this category are not comparable, because some include a real agent and some include autocomplete with an agent-shaped button.

VendorFree tierIs there a real agent in it?
CodexIncluded in ChatGPT FreeYes. Same agent, smaller allowance
GitHub Copilot2,000 completions a month, limited chat and agent usageLimited, on auto model selection only
CursorHobbyLimited agent usage
DevinFree plan, one memberYes, with a light quota and limited models
Claude CodeNoneNo. Pro is the entry point
Gemini Code AssistWithdrawn 18 June 2026No longer

Five scenarios, priced

Sticker prices do not answer the question people actually have. These do.

01

Trying agent coding, budget zero

Codex on ChatGPT Free, plus Copilot Free in your editor. Total $0. You will hit limits, and hitting them tells you what you would be paying for.

02

One engineer, agent coding most working days, one session at a time

Claude Pro at $20 or ChatGPT Plus at $20 covers this if your sessions are focused. If you are stopped more than once a week mid-task, Claude Max 5x at $100 is the tier that stops the interruptions. Adding Copilot Pro at $10 for in-editor completion is the cheapest genuine upgrade on this row.

SetupMonthly
Claude Pro alone$20
Claude Pro + Copilot Pro$30
Claude Max 5x + Copilot Pro$110
03

One engineer running two or three agents in parallel all day

This is the row where the arithmetic stops being obvious. Parallel sessions on one account draw from one allowance, so three agents exhaust it roughly three times as fast.

OptionMonthlyWhat you get
Claude Max 20x$200One large window
Two Claude Max 5x accounts$200Two independent windows
Claude Max 5x + ChatGPT Plus$120Two vendors, two allowances
Cursor Ultra$200One large credit pool

Two $100 subscriptions cost the same as one $200 subscription and cannot starve each other. Whether that is better depends on whether your constraint is one enormous session or several concurrent ones.

04

A five-person team

StackMonthly for 5
Copilot Business, $19/seat$95
Claude Team standard seats, $20/seat annual$100
Gemini Code Assist Standard, $19/seat annual$95
Cursor Teams, $40/seat$200
Devin Teams, 5 full seats at $40$200
Claude Team premium seats, $100/seat annual$500

Seat maths is where the cheap-looking vendors get expensive and the expensive-looking ones stay flat. Note that Devin Teams has a $80 a month floor regardless of seat count, and that its flex seats cost nothing but draw from a shared credit pool instead of carrying their own quota.

05

CI, scripts, or anything programmatic

None of these subscriptions is the right instrument. An interactive plan cannot hold a session inside a GitHub Action, and the terms are written for a person. Use API keys with a budget limit, and price the workload per token rather than per seat.

The comparisons that do not survive contact

  • Comparing $20 to $20. Claude Pro at $20 and Cursor Pro at $20 buy different units of a different thing. One is a usage window on a terminal agent, the other is a credit pool in an editor.
  • Ignoring what does not consume credits. On Copilot, completions and next edit suggestions are unlimited on paid plans and consume no credits at all, so the credit allowance only describes chat and agent work.
  • Pricing the model instead of the session. Agent sessions re-send accumulated context every turn, so the second hour of a session costs several times the first. Per-million rates understate long sessions badly unless you model caching.
  • Assuming a free tier is permanent. Gemini Code Assist for individuals is the counterexample, withdrawn on 18 June 2026 with users pointed at Antigravity and Antigravity CLI.
  • Counting only one subscription. Most people who do this seriously end up paying two vendors. The honest comparison is your total, not your cheapest line item.

Which one to buy

If this is youBuy
Evaluating, zero budgetCodex on ChatGPT Free
Verified studentCopilot Student, free, then add a small API balance
One focused agent session most daysClaude Pro $20 or ChatGPT Plus $20
Interrupted mid-task more than weeklyClaude Max 5x $100
Two or three parallel agentsTwo Max 5x accounts, not one Max 20x
You live in the editor, not the terminalCursor Pro $20, or Copilot Pro $10
A team that wants central billing and policyCopilot Business $19/seat or Claude Team seats
Already on Google CloudGemini Code Assist Standard, $19/seat annual
Delegating whole tickets asynchronouslyDevin Pro $20, then Teams if it sticks
CI and automationAPI keys with a budget limit, no subscription

Buy one tier below what you think you need. Upgrading takes a minute, every one of these plans is monthly, and the wasted month you avoid is real money. The signal to move up is not hitting a limit once, it is hitting limits during work you had planned to finish that day.

Questions people ask

Codex, at $0, because OpenAI includes it in every ChatGPT plan including Free. Among paid entry tiers, GitHub Copilot Pro is $10 a month, and Claude Code, Cursor, and Devin all start at $20.

Both start at $20 a month. Above that they diverge: Claude publishes $100 for Max 5x and $200 for Max 20x, while Cursor does not publish prices for Pro+ and Ultra, describing them only as 3x and 20x Pro limits. They are also different products, so the price is rarely the deciding factor.

No. Gemini Code Assist IDE extensions and Gemini CLI stopped serving the Code Assist for individuals, Google AI Pro, and Google AI Ultra tiers on 18 June 2026. Paid Standard and Enterprise licences and API-key authentication are unaffected.

Free, Pro at $20 a month, Max at $200, and Teams at an $80 monthly minimum with $40 full seats and free flex seats. The old ACU pricing has been replaced by plan quotas with metered overage at API list prices.

Many people pay for both, because they address different hours of the day. An editor assistant makes the time you spend in code faster; a terminal agent removes some of that time. It is waste only if your work has quietly consolidated onto one of them.

For parallel agents, usually yes. Two Max 5x accounts cost the same as one Max 20x and give you two independent usage windows that cannot starve each other. One big plan is better if your constraint is a single very heavy session.

None of the interactive subscriptions. A pipeline cannot hold a subscription login, and the terms are written for a person. Use API keys with a console budget limit for anything programmatic.

Often enough that undated comparisons are worthless. In the last year Gemini withdrew its free individual tier, Devin replaced its whole billing unit, and Windsurf was renamed. Check the vendor page before you budget, including against this one.

Sources

Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.

  1. Claude plans and pricing Pro, Max, Team, and Enterprise seat prices
  2. Codex and ChatGPT pricing plan prices, Codex inclusion, credit rates
  3. GitHub Copilot plans per-seat prices and AI credit allowances
  4. Gemini for Google Cloud pricing Code Assist Standard and Enterprise licence fees
Try it

The number the
pricing pages omit.

Continuum prices what you actually consumed across Claude Code, Codex, and the rest, by repo and model and day, so plan choice stops being a guess. Free app, your own subscriptions.

free app · your subscriptions · local-first