AI Coding Assistant Pricing 2026: What Eight Working Plans Cost

Eight coding assistants with a public working plan have a median price of $19.50 a month. Prices were read on each vendor's own page on October 17, 2026. For 10 developers on the entry team seat, GitHub Copilot Business and Amazon Q Developer Pro are the lowest matching bills, at $190 a month before overage.

An AI coding assistant suggests or edits code in an editor, a terminal, or a cloud agent. This page is the bill for those plans, and which assistant fits a workflow is covered in best AI coding assistants.

The cheap line on a pricing page is often a light tier. Codex Go and Cursor Plus both exist, and neither is the plan the vendor describes for regular agent work, so the cheapest row is the wrong line to budget. Budget the working plan, then read the meter that sits behind it, because that meter is what a heavy month adds.

Devshot data: a price book built from our own reviews already had a recorded line for each of these eight assistants. This page replaced those lines with a fresh read of the vendor pages on October 17, 2026, then took the median and the 10-developer bills from that read.

Toolradar data: of the 180 AI coding tools on the AI coding statistics page, as of October 16, 2026, the median entry paid plan is $19.50/mo for plans with no stated unit, across 62 tools. The eight-assistant median above lands on the same figure, from a narrower set. The wider market count, including news volume, is on AI coding statistics.

What does a working plan cost?

A working plan here is the cheapest paid plan on which the vendor includes the assistant for regular coding. Light tiers, model-restricted tiers, and editor-only tiers stay in the notes, and they do not set the median. Every figure is a list price before tax, read on October 17, 2026.

Assistant	Working plan	Monthly price	Free path	Highest listed individual plan
GitHub Copilot	Pro	$10/mo	Yes	Max, $100/mo
Cursor	Premium	$30/mo	Yes	Ultra, $200/mo
Claude Code	Pro	$20/mo, or $17/mo annual	Chat only, Claude Code excluded	Max 20x, $200/mo
Codex	Plus	$20/mo	Yes, plus a lighter Go tier at $8/mo	Three Pro tiers, from $100/mo up to $500/mo
Amazon Q Developer	Pro	$19/mo per user	Yes, 50 agentic requests a month	Only one paid plan
Zed	Pro	$10/mo, with $5 of tokens included	Yes, editor plus your own keys	Only one individual paid plan
Devin	Pro	$20/mo	Yes	Max, $200/mo
JetBrains AI	AI Pro, individual	$10 per 30 days	Yes, 3 credits per 30 days	AI Ultimate, $30 per 30 days

Ordered from low to high, the working-plan column is three plans at the bottom step, then Amazon Q, then three plans on one shared sticker, then Cursor. Eight prices have no single middle row, so the median is the midpoint of the fourth and fifth, which is the figure in the opening sentence.

A solo developer comparing stickers will see the same monthly number on Claude, Codex, and Devin, with Copilot, Zed, and JetBrains a step below, and Cursor's recommended tier a step above. Inside this set, that is the whole spread of published working plans, and AI-assisted coding is the workflow context for picking among them.

Claude Pro is the only individual plan in the table with a published annual rate, shown on that row, while Max is monthly only. Codex Plus has no annual price on the Codex pricing page, so a buyer who pays each month has no yearly discount to take. Cursor sells a year at five sixths of twelve monthly payments, for Plus through Ultra, which is a real cut if the year is paid up front. Copilot's plans page lists monthly prices and does not show an annual individual rate, so a solo Copilot buyer should budget the monthly line with no yearly break.

What does a team of 10 pay?

The scenario is 10 developers for one month, on each vendor's lowest team seat that includes the assistant, billed monthly where a monthly rate exists. Tax, overage, and add-on credits are left out of every row, so the table is the seat bill before a heavy month adds usage. Where a vendor has no cheaper team tier, the row uses that seat.

Assistant	Entry team seat	How it is billed	Bill for 10
GitHub Copilot	Business	$19 per granted seat	$190
Amazon Q Developer	Pro	$19 per user	$190
JetBrains AI	AI Pro, per-user organization license	Twice the individual AI Pro price, per 30 days	$200
Claude	Team Standard	$25 per member, billed monthly, 2 seats minimum	$250
Codex	Business	$25 per user, billed monthly, 2 users minimum	$250
Cursor	Teams, Premium seat	$30 per seat	$300
Zed	Business	$30 per seat, and the seat includes no model credits	$300
Devin	Teams, full seat	$40 per seat once two full seats are in place	$400
Augment Code	Standard	$20 flat for up to 50 seats	Flat fee, not multiplied by 10

Copilot and Amazon Q tie, because both are per-user meters with no seat minimum, so a team of 10 is ten times one seat. AWS's own billing example for 10 Amazon Q users lands on the same monthly total as that row. Claude Team and Codex Business share one monthly seat rate, and both require two members before the plan exists at all, so a solo developer cannot buy either seat.

Annual billing changes two of those rows and leaves the tie alone. Claude Team Standard and Codex Business are each a fifth lower when billed annually, while Copilot Business and Amazon Q have no annual discount on their pricing pages. Cursor's year is five sixths of twelve monthly payments, including the Premium seat a team buys. At 10 seats the Copilot and Amazon Q tie still holds after that annual cut, because the two vendors that discount do not start from the lowest monthly seat.

Devin's minimum is the line buyers misread, because every Teams account pays at least $80 a month, and below two full seats the gap is filled with prepaid on-demand credits. At two full seats the minimum is already met, and further full seats are the per-seat line only, so ten full seats do not pay the minimum on top.

Augment's bill for 10 people is the flat fee in that row, not ten seats, and it is the least like the others. One flat fee covers up to 50 seats, and the included usage equals that fee. Past that pool, model inference is the provider's API list price plus a 40% service fee, while compute is extra and does not carry that fee, so a team that actually codes will leave the included pool quickly. The Business tier is a larger flat pool at $100 a month, still capped at 50 seats, still with the same fee on LLM usage above the pool.

Zed's team row is an admin product, because Business seats do not bundle a model allowance. Hosted usage bills at API list price plus 10%, or the team brings keys and pays the provider directly, so that row is control-plane cost and the model spend sits on top of it.

The larger-team cut, with a different headcount and a yearly frame, is best AI coding assistants for enterprise, and Devshot's daily brief flags when these seat prices move. Join at devshot.co.

Where does the sticker stop being the bill?

GitHub Copilot

Paid plans keep code completions and next-edit suggestions off the credit meter, while chat, agent mode, code review, the cloud agent, and the CLI draw GitHub AI credits. One credit is $0.01, and Pro includes 1,500 credits in the month, split into 1,000 base credits and a 500-credit flex allotment. GitHub says the flex allotment can change, and when it does the base credits still match the subscription, while the flex portion is the part that moves.

Pro+ is $39/mo and includes 7,000 credits, and the Enterprise seat matches that individual price, with 3,900 credits pooled at the billing entity. Additional credits past the pool cost the same $0.01. A developer who mostly accepts completions can stay on Pro, while a developer who runs an agent on a frontier model for hours will empty the pool and should read GitHub Copilot pricing before treating the seat as a ceiling.

Students, teachers, and some open-source maintainers may qualify for a free Pro seat. The claim path is in how to get GitHub Copilot for free, and the catalog entry is GitHub Copilot on Toolradar.

Cursor

Premium is the working plan because Cursor's pricing FAQ points new AI users there, and Plus is $8/mo and runs Grok models only. Third-party models start on Premium, billed at API list price from the first request, and they are not inside the monthly fee, so the Premium sticker is a floor once a buyer leaves Grok. Super, at $100/mo, sits between Premium and Ultra for daily agent use.

Teams cannot buy Plus, so the cheapest team seat is Premium, then Super, then Ultra, and a team can mix them. On Teams and Enterprise, third-party requests also add a Cursor Token Rate of $0.25 per million tokens, and individual plans do not pay that rate, so a heavy solo user can avoid that meter by staying off the team plan.

The usage pool refreshes weekly, so a heavy day does not spend the whole month, and older subscriptions can keep renewing on earlier terms. The price a new buyer sees is the table, with the detail in Cursor pricing and the Cursor review, and the catalog entry at Cursor on Toolradar.

Claude Code

Claude Code is included on every paid Claude plan and excluded from Free, so a free account cannot run the terminal agent. Terminal sessions and chat draw one usage pool, with a five-hour window plus weekly limits on paid plans, which means a long coding session and a long chat spend the same allowance. Pro is the individual entry, Max 5x is five times Pro's per-session allowance, and Max 20x is twenty times, both monthly only.

The pricing page also attaches a monthly API credit equal to that Max subscription, which is platform credit on top of the plan, not a second seat.

Team Standard is 1.25 times Pro's per-session allowance, with its own weekly cap, and usage does not pool across teammates, so one quiet seat cannot rescue a teammate. Premium seats are five times the Standard monthly price and include 6.25 times Pro's per-session allowance, which is the step when the weekly cap is the limit. A company that wants the agent and the chat under one bill should start on Pro for a single developer, or on Team Standard when two people need admin controls.

Enterprise, on the pricing page, is a per-seat fee plus usage at API rates, billed annually, so there is no monthly Enterprise price for the ten-person table. The plan-by-plan walkthrough is Claude Code pricing, the product page on this site is the Claude Code review, and Claude Code vs Cursor is the workflow comparison.

Codex

Codex is included in ChatGPT Free, Go, Plus, Pro, Business, Edu, and Enterprise, and Go is the light paid tier rather than the plan for regular agent work. Plus is the working plan: Codex in the CLI, the IDE extension, on the web, and on iOS, with cloud review and Slack. Pro has three monthly tiers, the pricing page gives the floor and the ceiling in the table, and Ultrafast mode is only on the top tier, so that mode requires the ceiling tier.

Pro plans currently have no five-hour limit, while Plus and standard Business still do, and the estimate ranges on the pricing page move with the model, so a Plus user can hit a session wall that Pro does not.

Business is the team seat, two users minimum, and the page leads with the annual rate while this scenario uses the monthly rate from the same page, which is higher, so a month-to-month quote that uses the headline will come in low. Past the included allowance, Plus and Pro can buy credits. An API key is a separate meter, with no cloud review and no fixed seat, which suits a script that already holds a key.

ChatGPT vs Claude for coding compares the two subscriptions as coding tools, and OpenCode vs Claude Code covers a client that can sit on either provider.

Amazon Q Developer, Zed, Devin, JetBrains

Amazon Q's free tier allows 50 agentic requests a month in the IDE or CLI, and 1,000 lines of code a month for Java upgrades, which is a trial rather than a month of upgrade work. Pro raises the Java allowance to 4,000 lines per user, pooled at the payer account, then charges $0.003 per extra line submitted, so price a large upgrade in lines before the month starts. Signing in does not start the subscription: the fee starts when someone requests a plan, runs a transformation, or accepts a completion, and a mid-month cancel still bills the full month. There is no second paid tier, so a team that outgrows Pro pays the per-line charge on the same seat, and the Toolradar profile lists the product.

Zed Pro is an editor subscription with a small hosted balance, then usage at API list price plus 10%, so the sticker buys the editor plus a thin balance, not a month of hosted models. The 14-day trial is a $5 balance of GPT-6 Luna only, shared with Delta, and it ends when the balance or the 14 days run out. Personal stays free if you bring keys or an external agent, and Business adds org controls without adding a credit pool, so that seat is administration, not tokens. The catalog entry for that editor is Zed on Toolradar.

Devin Pro is a single user, with a daily and weekly quota across the cloud agent, the CLI, and the desktop, so the daily cap can stop a session while the week still has room. Max drops the daily cap and raises the weekly quota, which is the plan for consecutive heavy days. Teams full seats include a Pro-equivalent quota and the desktop app, while Flex seats are free, draw only on shared on-demand credits, and do not include the desktop. Purchased credits roll over and do not expire, and public pull-request review on Devin's review site does not consume credits. The catalog entry for the agent is Devin on Toolradar.

JetBrains prices individual use and organization licenses on separate tabs of the same help page, and individual AI Pro is the working plan in the first table. The organization tab is twice that price for AI Pro, and the credits scale with the price, with one credit equal to $1. Ultimate adds a small bonus above the price, and bought credits last 12 months. Since July 13, 2026 the hosted service is unavailable in China, while API keys and local models still run there, so a team there cannot buy the hosted assistant. AI Free cannot top up, which means a free user has to move onto a paid plan once the included credits are gone.

Augment

Augment does not sell a per-developer seat on Standard or Business: both are one monthly fee, up to 50 people, with a shared usage pool equal to the fee, so headcount inside that cap does not change the sticker. The 40% service fee applies to LLM tokens and not to Cosmos compute, and top-ups that are not part of the subscription last 12 months, which suits a team that bursts and then goes quiet. Enterprise is custom, and it is the tier that removes the 50-seat cap, so past that cap there is no self-serve price. A five-person startup and a 40-person org pay the same Standard sticker, which is the opposite of every per-seat row above, until the pool runs out. The catalog entry for that flat fee is Augment Code on Toolradar.

What is missing from this price set?

Gemini Code Assist is not in the tables, because Google's documentation says new subscriptions are no longer available for purchase, so there is no current self-serve seat to put in a 2026 bill. Tabnine publishes no list price, and its company page now presents Tabnine as part of Tricentis and offers a sales conversation, with no plan amounts, so a buyer cannot place it against the median until a quote arrives.

Bring-your-own-key editors, where the client is free and the model bill is yours, are outside this median on purpose, because their cost is a token invoice, not a seat. That comparison is best free AI coding tools, and app-builder pricing, which is a different purchase, is not in this set either.

How we measured

The working-plan rule and the 10-developer rule are above. Prices were read on October 17, 2026 on GitHub Copilot plans, GitHub's Copilot plan docs, Cursor pricing, Cursor model pricing, Claude pricing, Claude's Max plan article, Claude's Team plan article, Codex pricing, Amazon Q Developer pricing, Zed pricing, Devin's self-serve plan docs, JetBrains AI licensing, and Augment pricing. The closed-subscription note is from Gemini Code Assist's overview. The missing Tabnine amounts are from Tabnine's pricing page. The catalog median is from Toolradar's AI coding statistics, dated October 16, 2026 on that page.

Head-to-head fit, rather than the bill, is in Cursor vs GitHub Copilot and the GitHub Copilot review. This page is recalculated when the vendor pages change, and the updated date moves with that read.

FAQ

What is the median AI coding assistant price in 2026?

Across the eight working plans in the table, the median is $19.50 a month, the midpoint of Amazon Q Developer Pro and the shared sticker on Claude, Codex, and Devin. It matches Toolradar's median entry price for AI coding tools that publish a plan with no stated unit, a wider and older snapshot from October 16, 2026. This eight-assistant set is not a cheap slice of that wider catalog.

What is the cheapest AI coding assistant in 2026?

Three working plans are $10: GitHub Copilot Pro, Zed Pro, and JetBrains AI Pro for an individual, and they are not the same purchase. Zed's price includes a small token balance and then meters hosted models, so the sticker runs out once hosted models are in use. Copilot's price includes unlimited completions on the paid plan and a credit pool for agent use, which suits someone who mostly accepts suggestions. JetBrains bills that amount per 30 days and includes 10 credits, with each credit equal to $1 of usage, and a free path exists on all three with tighter limits than the paid plan.

How much does a team of 10 pay for GitHub Copilot?

Copilot Business is one granted seat times 10, which is the $190 row, and completions stay off the credit meter while agent, chat, and the CLI spend the included credits. Extra credits bill at the same credit rate, so a team that lives in agent mode should not treat that row as a ceiling. Enterprise is the higher governance tier and matches the Pro+ seat price, which buys org controls rather than a cheaper agent.

Is Claude Code included in Claude Pro?

Yes. Claude Code is on every paid Claude plan, including Pro, both Max tiers, and Team, and it is not on the Free plan, so a free account has to upgrade before the terminal agent exists. Usage is shared with chat, so a long terminal session spends the same pool as a long conversation. Max is monthly only, while Pro is the plan with an annual option, which makes Pro the individual plan when a yearly bill matters and Max the plan when the allowance is the constraint.

Is there a free AI coding assistant?

Yes, with caps. Copilot, Cursor, Codex, Devin, Amazon Q, Zed, and JetBrains each publish a free path, while Claude's free plan does not include Claude Code. Amazon Q's free path is 50 agentic requests a month, and Copilot Free caps completions at 2,000 a month. None of those free paths is the working plan in the median, so a team budgeting real use should start from the paid rows. Quota-level comparisons are in best free AI coding tools.

Why is Augment's flat fee not the cheapest team plan?

The fee covers up to 50 seats, and the included usage equals the fee, so the sticker looks cheap next to a per-seat plan only while the team stays inside that pool. LLM usage past that pool adds a 40% service fee on the provider's API price. A team that stays inside the included pool can come in under the per-seat rows, and a team that runs agents all week will pay the top-ups until the sticker stops being the bill.

Cite this: Devshot, "AI Coding Assistant Pricing 2026: What Eight Working Plans Cost", October 2026.

---

Devshot covers the pricing and product changes developers have to act on, in one daily brief. Join at devshot.co.

Some offers on this page may be paid placements or contain affiliate links.