Key findings
- Team seats billed monthly list per user at GitHub Copilot Business US$19. Gemini Code Assist Standard lists at US$22.80, but Google has sold no new Code Assist subscriptions since October 9, 2026. Claude Team Standard and ChatGPT Business Standard (with Codex) seats cost US$25, and Devin Teams costs US$40. Cursor Teams seats start at US$30 in the X Subscription Cursor announced October 12, 2026; Standard seats at US$40 remain only for teams created before it.
- Seats carry a usage allowance that heavy agent work exceeds. Claude plans reset on five-hour and weekly windows, Copilot pools 1,900 or 3,900 AI credits per seat at US$0.01 each, and Cursor's X Subscription plans refresh one weekly pool for Cursor's own models and bill third-party models at API list price from prepaid credit.
- Anthropic reports that enterprise Claude Code deployments average about US$13 per developer per active day and US$150 to US$250 per month, with 90% of users under US$30 per active day.
- Claude Code Review averages US$15 to US$25 per review in Anthropic's launch figures, billed on tokens through usage credits under a monthly spend limit. GitHub estimates US$0.05 to US$1 per Copilot review on Lite effort and US$0.25 to US$5 on Balanced.
The seat is the smallest cost of an accepted change. Engineer time to review and rework the change costs more than the seat, the metered usage, or the automated review. Every vendor that sells a seat caps or meters model usage on top of it, and automated review bills on a meter of its own. GILD compares list prices for seven vendors, and sets out a method for cost per accepted change.
List prices, side by side
Every product below except Cline sells a team seat. Annual billing lowers the Claude Team Standard seat and the ChatGPT Business Standard seat to US$20 per user per month, and Gemini Code Assist Standard to US$19 on a 12-month commitment, for existing subscribers only: Google stopped selling new Code Assist subscriptions on October 9, 2026. Cline's open-source agent has no seat fee, so you pay the model provider directly or buy Cline credits.
| Product | Team seat, billed monthly | Individual plans | Team plan | Enterprise |
|---|---|---|---|---|
| Claude Code (Claude plans) | Pro US$20 per month or US$17 billed annually; Max from US$100 per month | Team Standard US$25 per seat per month, or US$20 billed annually; Premium US$125, or US$100 annually | US$20 per seat per month billed annually, plus usage at API rates | |
| Codex (ChatGPT plans) | Go US$8 and Plus US$20 per month; Pro tiers at US$100, US$200, or US$500 per month | Business US$25 per user per month, or US$20 billed annually for 2+ users; Premium US$125, or US$100 annually | Contact sales | |
| Cursor | Plus US$8 (Grok models only), Premium US$30, Super US$100, Ultra US$200 per month in the X Subscription since October 12, 2026; Pro US$20 and Pro+ US$60 are legacy plans closed to new customers | Teams seats at US$30 (Premium), US$100 (Super), or US$200 (Ultra) per user per month; Standard US$40 and Premium US$120 are legacy seats for teams created before the X Subscription | Custom | |
| GitHub Copilot | Pro US$10, Pro+ US$39, Max US$100 per month | Business US$19 per granted seat per month | US$39 per granted seat per month | |
| Devin and Devin Desktop (formerly Windsurf) | Pro US$20, Max US$200 per month | Teams US$40 per full seat per month, US$80 monthly minimum, flex seats at no separate charge | Contact sales | |
| Gemini CLI (Gemini Code Assist) | Paid Gemini API key at per-token rates; the free tier moved to Antigravity CLI on June 18, 2026 | Standard US$22.80 per user per month on a monthly commitment, US$19 on a 12-month commitment, for existing subscribers only: new Code Assist subscriptions closed on October 9, 2026 | Enterprise US$54, or US$45 on a 12-month commitment, for existing subscribers only | |
| Cline | Free open-source agent; ClinePass US$9.99 per month standard rate | No seat fee; bring your own provider key or buy Cline credits | Custom |
Vendor pricing pages listed under Sources, checked October 12, 2026; Cursor, GitHub Copilot and Gemini Code Assist rechecked October 28, 2026. US dollars, before tax. Gemini Code Assist prices are Google's published hourly license rates over a 730-hour month.
What the seat includes, and what happens past it
Each vendor caps the agent usage a seat includes and charges for usage past the cap. Cursor's seats include no third-party model usage at all, which bills at API list price from the first request. Two teams on the same seat can therefore pay very different bills. Check who can turn on paid usage and whether a limit stops work or only sends an alert. GitHub turns paid usage on by default for organizations until an administrator disables it.
| Product | Included in the seat | Past the allowance |
|---|---|---|
| Claude Code | A per-seat allowance shared with Claude chat and Cowork, reset on a five-hour and a weekly window; Premium seats get more | Usage credits, with spend limits for the organization, a group, or a member |
| Codex | Plus and Business: 15 to 160 local messages per five hours on GPT-6.1 Sol; Pro has no five-hour limit; weekly limits may apply | Credits on Plus and Pro, workspace credits on flexible-pricing plans, or an API key at API rates |
| Cursor | One weekly pool for Cursor's own models per seat, shared with Grok; Super seats get 3.75 times and Ultra seats 12.5 times the Premium pool, and third-party models are not included. Legacy Standard and Premium seats keep two monthly pools, Cursor Models and Other Models | Prepaid team balance: third-party models at API list price plus US$0.25 per million tokens, Cursor models past the pool at listed rates, under a monthly additional usage limit and per-member limits; legacy seats bill on-demand usage in arrears |
| GitHub Copilot | 1,900 (Business) or 3,900 (Enterprise) AI credits per seat per month, pooled across the organization; 1 credit is US$0.01 | Paid usage, on by default; administrators disable it or set user, cost-center, or enterprise budgets |
| Devin and Devin Desktop | A daily and weekly quota on Pro; a larger weekly quota with no daily cap on Max | Prepaid on-demand credits that roll over and never expire, shared across a Teams plan |
| Gemini CLI | 1,500 (Standard) or 2,000 (Enterprise) model requests per user per day | A paid Gemini API key at per-token rates |
| Cline | No allowance with your own key; ClinePass quotas give 2 to 5 times standard API rate limits | The model provider's rates, or pay-as-you-go Cline credits |
Vendor documentation listed under Sources, checked October 12, 2026; Cursor rechecked October 28, 2026.
What heavy use costs
Anthropic publishes the clearest usage figure. Across enterprise deployments, Claude Code averages about US$13 per developer per active day and US$150 to US$250 per developer per month, and 90% of users stay below US$30 per active day. Heavy individual use runs far higher. A Sanity Staff Engineer (at the time of writing) budgets US$1,000 to US$1,500 a month for a senior engineer working mostly through the agent.
Token rates set the floor for metered and API use. Per million tokens, Claude Opus 5.5 costs US$4 input and US$20 output and Sonnet 5.5 US$2 and US$10; OpenAI's gpt-6.1-sol costs US$2 and US$10. Anthropic names the usual causes of unexpected spend as long sessions that were never cleared and Opus left as the default model. It reports that agent teams use about seven times the tokens of a standard session when teammates run in plan mode.
| Source | Basis | Per developer per month |
|---|---|---|
| Anthropic, enterprise average (low end) | Vendor-reported across enterprise deployments | |
| Anthropic, enterprise average (high end) | Vendor-reported across enterprise deployments | |
| Sanity Staff Engineer at the time, heavy use (low end) | Self-reported budget, September 2025 | |
| Sanity Staff Engineer at the time, heavy use (high end) | Self-reported budget, September 2025 |
Claude Code docs: Manage costs effectively, checked October 12, 2026; Sanity, September 2, 2025.
Automated review is a second meter
AI review cost is the cost of each review times the number of reviews, and the vendors meter both differently. Claude Code Review runs several agents on Anthropic's infrastructure and bills the organization for tokens through usage credits, outside plan usage. It has no flat price, and Anthropic's launch post puts the average at US$15 to US$25 a review. Each review costs more with a higher review level, a larger change, more complex surrounding code, and more candidate issues to verify. GitHub publishes an estimated AI credit range per Copilot review, plus GitHub Actions minutes. It bills automatic reviews to the pull request author and manual ones to whoever requests them, and an owner can bill the organization instead. Cursor moved Bugbot to usage-based billing in its May 2026 pricing update.
Your own repositories decide where Claude's cost falls in that range, and Anthropic's documentation gives the method. Set a monthly Code Review spend limit and turn review on for a few representative repositories, set to run once per pull request or on request. Read each repository's average cost per review in the admin settings, then multiply it by the pull requests that repository opens in a month. The spend limit applies to the whole organization. When it, the organization's overall limit, or the usage credit balance runs out, Code Review skips the review and posts a comment on the pull request naming the cause.
Pick the trigger deliberately. A repository added to Claude Code Review from the admin page starts on every push, the most expensive setting, which multiplies the cost by the number of pushes. Once per pull request is cheaper. Per-push review suits repositories where a missed defect costs more than the reviews.
| Reviewer | High end per review | Billed to | Published cost per review |
|---|---|---|---|
| Claude Code Review | The organization, on tokens through usage credits, under a monthly spend limit for the whole organization | No flat price; each repository's average cost appears in admin settings after reviews run | |
| Copilot code review, Balanced effort | The pull request author for automatic reviews, or the user who requests one, in AI credits, plus GitHub Actions minutes; owners can bill the organization instead | US$0.25 to US$5 | |
| Copilot code review, Lite effort | The pull request author for automatic reviews, or the user who requests one, in AI credits, plus GitHub Actions minutes; owners can bill the organization instead | US$0.05 to US$1 |
Claude Code docs: Code Review costs and usage, checked October 25, 2026. Claude blog: Code Review for Claude Code, March 9, 2026, for the average. GitHub Copilot docs: Copilot code review, checked October 28, 2026. GitHub's figures exclude Actions minutes.
Cost per accepted change
Add four costs for a month, then divide by the changes that merged after normal review. Engineer time usually dominates. METR estimated 26 to 42 minutes of engineer time to bring a 2025 agent pull request to mergeable, and Faros AI measured review time up 91% on high-adoption teams. At typical loaded engineering rates, that time outweighs a US$13 day of tokens.
- Seats: the monthly seat price for every engineer with access, used or not.
- Metered usage: usage credits, overage, on-demand spend, or API bills for the month.
- Automated review: cost per review times reviews per pull request, which depends on the trigger. For Claude Code Review, start from Anthropic's published average and replace it with your pilot's per-repository figure.
- Engineer time: review and rework hours on agent-assisted changes, at your loaded hourly cost.
Controls that keep spend predictable
Set the caps before rollout, then watch per-user spend weekly for the first month.
- Claude Team and Enterprise: turn on usage credits only with spend limits at the organization, group, or member level. On API billing, set Console workspace spend limits and export per-user cost through OpenTelemetry.
- GitHub Copilot: decide whether the AI credits paid usage policy stays on, and set user-level budgets; a US$0 budget blocks a user immediately.
- Cursor and Devin: usage past the included pool bills from prepaid credit or top-ups, or as on-demand usage on legacy Cursor plans, so assign an owner for the monthly total.
- Gemini Code Assist, for existing subscribers: the seat sets a daily request cap, and work past it moves to an API key, so check which keys engineers use.
Methodology
List prices are each vendor's own, in US dollars before tax, from the official pricing pages and documentation GILD checked on October 12, 2026. GILD checked Claude Code Review billing on October 25, 2026, and rechecked Cursor, GitHub Copilot and Gemini Code Assist on October 28, 2026. Anthropic reports its own usage costs, Sanity and incident.io report theirs, and review time figures come from METR and Faros AI. The cost-per-accepted-change method is GILD's.
Products in this piece
- Claude Code+75 Strong positiveAI Coding Agent by Anthropic
- Codex+55 Strong positiveAI Coding Agent by OpenAI
- Cursor+11 MixedAI Coding Agent by Anysphere
- GitHub Copilot+15 MixedAI Coding Agent by GitHub
- Gemini CLI−3 MixedAI Coding Agent by Google
- Devin+39 PositiveAI Coding Agent by Cognition
- Devin Desktop (formerly Windsurf)+17 MixedAI Coding Agent, also by Cognition
- Cline+39 PositiveAI Coding Agent by Cline
Sources
- Official documentation
- 22
- Public source
- 4
- Claude pricing
- Claude Code docs: Manage costs effectively
- Claude API pricing
- Claude Code docs: Code Review
- Claude Code docs: Code Review costs and usage
- Claude blog: Code Review for Claude Code
- Codex docs: Pricing
- ChatGPT Business pricing
- OpenAI API pricing
- Cursor pricing
- Cursor forum: Introducing The X Subscription, staff announcement (Oct 12, 2026)
- Cursor docs: Models & pricing
- Cursor docs: Bugbot
- GitHub Copilot docs: Plans
- GitHub Copilot docs: Usage-based billing for organizations and enterprises
- GitHub Copilot docs: Copilot code review
- Devin docs: Self-serve plans
- Gemini for Google Cloud pricing
- Gemini Code Assist docs: Gemini Code Assist Standard and Enterprise subscriptions
- Gemini CLI docs: Quota and pricing
- Cline pricing
- ClinePass
- Sanity: First attempt will be 95% garbage (Sep 2, 2025)
- incident.io: How we're shipping faster with Claude Code and Git Worktrees (Jun 27, 2025)
- METR: Research Update, Algorithmic vs. Holistic Evaluation (Aug 2025)
- Faros AI: The AI Productivity Paradox Research Report (Jul 23, 2025)