AI coding agent pricing in 2026: what Claude Code, Codex, Devin and Copilot cost

Published by 3Gen Internet Corporation, maker of OpenAssistant, Skej and AGNT. Comparisons use public documentation, without hands-on testing.

AI coding agent pricing is mostly a monthly plan with a usage allowance, not a price per task. On the vendors’ own pages, one developer pays $8 to $20 a month at the entry tier and $100 to $200 a month at the heavy-use tier (Codex Pro also has a $500 tier). Entry-level team seats run from $19 (Copilot Business) to $40 (Devin Teams full seat), before extra usage. The traps are in how each vendor counts usage, so read the limits column below before you compare sticker prices. For contrast, OpenAssistant is a general work assistant that charges in credits per task, covered near the end.

The five pricing patterns

The vendor pages show five ways to charge. Most products mix two or three.

1. Flat subscription tiers with usage limits. You pay a fixed monthly price and get an allowance that resets. Claude Code is included in paid Claude plans with limits that reset on a rolling five-hour window (Claude pricing). Codex is included in ChatGPT plans from Free to Enterprise (Codex pricing). Jules counts tasks per rolling 24 hours (Jules usage limits).
2. Credits. The plan includes a pool of credits, and agent work draws it down. GitHub states “1 AI credit = $0.01 USD.” (Copilot plans). Replit says each plan includes monthly credits, “$20 for Core, starting at $100 for Pro” (Replit pricing). Codex also lists credits per million tokens by model (Codex pricing).
3. Per seat plus usage. A seat fee covers access, and heavier work is billed at API rates. Claude Enterprise is $20 per seat per month billed annually, with a 20-seat minimum, and “Usage cost scales with model and task” (Claude pricing). Devin Pro and Teams let you buy extra usage “at API pricing” (Devin pricing).
4. ACUs or equivalent compute units. Devin Enterprise is billed in Agent Compute Units at the rate in the customer’s order form, and no public ACU price is shown (Devin enterprise billing). Devin’s older ACU plans were migrated: Legacy Core plan users moved to the Free plan (Devin self-serve billing). Cursor’s cloud agents are charged at API pricing for the chosen model (Cursor cloud agent docs).
5. Bundled plans. The agent is a feature of a larger subscription. Cursor “is sold as part of The X Subscription” (Cursor pricing docs). Jules paid limits come through Google AI plans, not from Jules itself (Jules usage limits). Codex and ChatGPT Work share usage (Codex pricing).

Every plan and price on the vendor pages

The table lists each product in order of its lowest paid individual plan. It is not a quality ranking, and sticker prices are not comparable because each vendor counts usage in a different unit. Prices are USD as the vendor shows them, excluding tax. “Not stated” means the vendor pages we read do not say.

Replit’s pricing page also lists higher Pro credit tiers at $215, $425, $825 and $2,000 a month, but does not label the credit amounts for the upper tiers (Replit pricing). Replit also documents a free Starter plan with daily Agent credits up to a monthly cap, although no Starter card appears on its pricing page (Replit Starter docs).

What limits the vendors state

Each vendor measures usage differently, so a $20 plan on one product is not the same as a $20 plan on another.

• Claude Code. “Every plan has usage limits that reset on a rolling five-hour session window.” Cloud sessions share rate limits with all other Claude and Claude Code usage (Claude pricing; Claude Code on the web docs). No numeric allowance per plan is published, only multipliers (5x, 20x).
• Codex. On Plus and standard Business, local messages per five-hour window are estimated at 15 to 160 for GPT-6.1 Sol, 350 to 3,000 for GPT-6 Luna and 5 to 45 for GPT-6 Astra. The vendor adds “These estimates are not fixed message limits” and says cloud tasks may use more of your allowance (Codex pricing). Plus and Pro users who hit a limit can buy more credits.
• Devin. Pro and Teams full seats get an allowance that refreshes daily and weekly. Max has a weekly allowance with no daily cap. The exact quota sizes are not published; the page says only “light” for Free and “significantly higher” for Max. Purchased credits never expire (Devin pricing; self-serve billing).
• GitHub Copilot. Credits are dollars at $0.01 each. Code completions and next edit suggestions do not use credits, while agent mode, code review, the cloud agent and Copilot CLI do (Copilot plans). You can set a dollar budget for extra usage, with alerts at 75%, 90% and 100%.
• Cursor. One pool of usage, shared across Cursor and Grok, refreshes weekly. Third-party model usage is not included in the monthly prices shown (Cursor pricing). The pool size in dollars or tokens is not published on the pages we read.
• Replit. Credits reset on the monthly renewal date, and extra usage is charged at pay-as-you-go rates up to a budget you set (Replit pricing). Pro credits roll over for two months (Replit Pro docs).
• Jules. Task counts per rolling 24 hours. Over the limit, the new task button is disabled. The vendor says “We may adjust limits and features as we learn” (Jules usage limits).

Traps to check before you buy

Devin Teams: the $80 minimum. The pricing page reads “$80 /month for team plan + $40/mo per full dev seat” (Devin pricing). That wording suggests $80 plus $40 for every seat. The billing docs read differently: “every Teams account pays at least $80/month,” with the first two full seats inside that $80, so 3 full seats cost $120 and N seats of 2 or more cost N x $40 (Devin self-serve billing). We follow the docs table because it gives worked numbers. The two readings differ by $80 a month for a five-seat team (see the example). Flex seats are free but draw only from the team’s shared credit pool and do not include Devin Desktop.

Cursor: bundled pricing and unclear meters. Cursor “is sold as part of The X Subscription,” and its pool is shared across Cursor and Grok (Cursor pricing docs). The $8 Plus plan allows only Grok models and no Cloud Agents. Third-party models are billed outside the listed prices, and Teams and Enterprise add a Cursor Token Rate of $0.25 per million tokens. The pricing page lists cloud agents as included from Premium, the cloud agent docs say they are “charged at API pricing for the selected model” (Cursor cloud agent docs), and the pricing docs say the weekly pool covers Cloud Agents. The pages do not agree on which meter applies.

Jules: prices are not on its pages. Jules lists task limits for Free, Pro and Ultra, but says the paid tiers are a benefit of the Google AI Pro and Google AI Ultra plans. No Jules page we read gives those plans’ prices (Jules usage limits). Paid limits are currently for individual Google Accounts only, and family plan limits are not pooled, so there is no team purchase path on the page. The Jules FAQ says it is in Public Beta (Jules FAQ).

GitHub Pro and third-party agents. The Copilot features page lists “Access to 3rd party agents (Claude Code and Codex)” under Pro (Copilot). The plans page table shows delegating tasks to third-party agents as not included on Pro and included on Pro+ and Max, marked Preview (Copilot plans). We follow the plans page. If you are buying Pro for that feature, check the live plans table first. GitHub also says code review workflows consume GitHub Actions minutes from 1 June 2026 and that flex credit allotments on Max “may change over time.”

Other things that change the bill.

• Codex’s own product page shows only Plus, Pro from $100 and Business, while the pricing doc also lists Free, Go and the $200 and $500 Pro tiers (Codex page). The product page note “Unlimited subject to abuse guardrails” is not explained on the pricing doc.
• Claude Max is billed monthly only, and annual billing exists for Pro and Team (Claude pricing). Payments are generally non-refundable.
• Claude Enterprise requires 20 seats at $20 annually, so a team of 5 cannot use that line.
• Devin Free shows no cloud sessions on the pricing page comparison, while the docs say Free includes access to DeepWiki and Devin Review. The pricing page cross for DeepWiki conflicts with the docs.

Worked example: one developer and a team of five

These numbers use only the list prices above, excluding tax and extra usage. They show the cost of access, not the cost of finishing a given amount of work, because the vendors do not publish comparable allowances.

Solo developer

Annual cost is the monthly price times 12, or the stated annual price where the vendor shows one.

At the heavy-use tier: $100 x 12 = $1,200 a year for Claude Max 5x, Copilot Max, the lowest Codex Pro tier, or Cursor Super (which is $1,000 on annual billing). Devin Max, Claude Max 20x and Cursor Ultra are $200 x 12 = $2,400 a year (Cursor Ultra is $2,000 on annual billing). Copilot Pro includes $15 of credits for $10, which is a ratio of 15 / 10 = 1.5. Copilot Pro+ is 70 / 39, about 1.8, and Max is 200 / 100 = 2.0, so higher tiers include more credit per dollar paid.

Team of five developers

Same product order as the plan table above. One seat per developer, no flex seats, no extra usage.

Three things stand out from the arithmetic. First, the two Devin readings differ by $280 - $200 = $80 a month, or $960 a year, so confirm with the vendor before budgeting. Second, annual billing cuts Codex Business and Claude Team standard by 5 x ($25 - $20) = $25 a month, or $300 a year. Third, a team of five cannot buy Claude Enterprise: 20 seats x $20 = $400 a month is the minimum, plus usage at API rates.

OpenAssistant: a credit model for contrast

OpenAssistant is not a specialist coding agent, and a coding specialist from the table above wins for pull-request work. We include it because its pricing shows a different model: credits per task rather than a coding allowance. A credit is “the unit of work your assistant uses to complete tasks,” with simple tasks at about 10 to 50 credits and deep research or multi-person scheduling at hundreds (OpenAssistant pricing details).

• Free Starter plan: 1,000 free credits at signup, then 100 bonus credits a day once used up, up to 1,000 a month. No credit card required (OpenAssistant pricing).
• Pro: $72 per user per month, or $60 on annual billing, at the 5,000-credit tier (OpenAssistant pricing; per-user pricing confirmed by 3Gen).
• Team: $120 per user per month, or $100 on annual billing, at the 5,000-credit tier (OpenAssistant pricing; per-user pricing confirmed by 3Gen).
• Higher credit tiers cost more. The pricing details page says Pro and Team both let you pick a credit tier from 5,000 to 100,000 credits a month, with the price scaling by tier. Enterprise is custom.
• Plan credits do not roll over from one billing cycle to the next.

For a five-person team on Team at the lowest tier, the arithmetic is 5 x $120 = $600 a month monthly or 5 x $100 = $500 a month annual. That is not comparable to the coding plans, since credits buy a different kind of work: scheduling, email, research and similar tasks across email, text, calls and Slack. For a broader view of credit, seat and per-resolution pricing across agent types, see AI agent pricing explained. For the coding specialists compared on capability, see best AI coding agents in 2026.

Related reading

• AI agent pricing in 2026: credits, seats and per-resolution fees explained: Credit plans, tasks, per-resolution fees, subscriptions and enterprise fees, with published prices and a worked example.
• Best AI coding agents in 2026: Claude Code, Codex, Devin and more compared: Seven coding agents compared on plan prices and data terms: Claude Code, Codex, Copilot, Devin, Cursor, Replit and Jules.
• How to choose an AI agent: 10 questions to ask before you buy: Ten questions on job fit, channels, approvals, price, data and exit, with a job-to-product table.
• Best free AI agents in 2026: what you get at no cost: True free plans, trials and expiring credits, with the stated limits and what each free tier lacks.
• Best AI agents in 2026: top picks by job, with prices: Our recommendations for each job (general work, coding, browser use, research, sales, support, enterprise and builders), with vendor prices.

Frequently asked questions

How much does Claude Code cost?

Claude Code is included in paid Claude plans and is not in the Free plan. Pro is $20 monthly or $17 a month on annual billing ($200 up front). Max is $100 for 5x Pro usage and $200 for 20x, billed monthly. Team standard seats are $25 monthly or $20 annually (Claude pricing).

How much does Devin cost?

Devin has a Free plan, Pro at $20 a month, Max at $200 a month, and Teams at an $80 monthly minimum with $40 per full seat. Enterprise is custom and billed in ACUs (Devin pricing). Extra usage on Pro and Teams is billed at API pricing.

Is there a free AI coding agent?

Several vendors list a $0 plan: Codex Free, Cursor Free, Copilot Free (2,000 completions a month), Devin Free (light quota, no cloud sessions per its comparison table) and Jules Free (15 tasks a day). Claude Code has no free plan. Limits are small, and vendors can change them.

What is the cheapest plan that includes a cloud agent?

On the pages we read, Copilot Pro at $10 a month lists the cloud agent as included (Copilot plans). Codex Go costs $8, but the page lists Go only with the desktop app, so which cloud features it includes is not stated. Cursor Plus at $8 excludes Cloud Agents, and Premium at $30 adds them.

How often do usage limits reset?

It depends on the vendor: a rolling five-hour window for Claude plans, a five-hour window for Codex local messages, daily and weekly allowances for Devin, weekly for Cursor, monthly for Replit credits and Copilot credits, and a rolling 24 hours for Jules tasks.

Try OpenAssistant

If you want an assistant that handles email, scheduling and follow-ups instead of writing code, OpenAssistant has a free Starter plan with 1,000 credits at signup and no credit card. See OpenAssistant pricing for the Pro and Team plans.