For 50 developers, Copilot has the lowest seat cost ($1,010 to $1,550 a month in our scenarios), but Claude Code and Codex seats are cheaper once developers run agents most days. Claude Code and Codex cost the same per seat: $1,240 a month for a mostly light team and $3,400 when 30 developers need premium seats. Copilot's included AI credits are worth its seat price, and agent work beyond them is billed at per-token rates. The same work that fits in Claude or Codex seats would cost $12,350 a month on Copilot for a mixed team. Run your own team in the calculator.
What do Codex, Claude Code and Copilot charge per seat?
| Standard seat | Heavy-use seat | Past the allowance | |
|---|---|---|---|
| OpenAI Codex (ChatGPT Business) | $20 annual, $25 monthly | Premium: $100 annual, $125 monthly; no 5-hour limit | Workspace credits, priced by your plan or agreement; or an API key at API rates |
| Claude Code (Claude Team) | $20 annual, $25 monthly | Premium: $100 annual, $125 monthly; 5x Standard usage | Usage credits at API rates, with limits per organization and member |
| GitHub Copilot | Business: $19; 1,900 AI credits | Enterprise: $39; 3,900 AI credits; needs GitHub Enterprise Cloud | AI credits at $0.01 each, priced per token for each model |
- Codex: included in ChatGPT Business, for teams of 2 to 200, with mixed seat types. OpenAI says that from December, plan usage will share one limit across ChatGPT and Codex. Sources: OpenAI business pricing and Codex pricing.
- Claude Code: each seat's allowance resets every 5 hours and every week, and is shared with Claude chat. The Team plan is for 2 to 150 people. Source: Claude pricing.
- Copilot: included credits are pooled across the organization and reset monthly. Each token is priced at the model's rate and converted to credits. For Claude Sonnet 5.5 ($2 in, $10 out per million tokens) and GPT-6.1 Sol, those rates match the providers' API list prices. Code completions don't use credits. Sources: Copilot plans and models and pricing.
The key difference: on Claude Code and Codex, a seat includes a usage allowance, and neither vendor publishes its size in dollars. On Copilot, the included credits are worth exactly the seat price at per-token rates. So Copilot charges agent work at roughly API rates, while Claude Code and Codex seats can include far more agent work than their price would buy.
How we built the three scenarios
The scenarios are illustrative, not customer data. We measure agent work in dollars at Claude API rates and use the same three developer profiles as our Claude Code seats vs API analysis. Two come from Anthropic's published averages.
| Profile | Active days | Per active day | Per month | Basis |
|---|---|---|---|---|
| Occasional | 4 | $5 | $20 | Illustrative light user: a few agent tasks a week |
| Typical | 15 | $13 | $195 | Anthropic's average per active day |
| Heavy agent user | 21 | $30 | $630 | The 90% ceiling, every working day |
- Seats: occasional and typical developers get the standard seat, and heavy agent users get the heavy-use seat. Prices are annual where offered. Copilot has one price. Copilot Enterprise needs GitHub Enterprise Cloud. Above the credit pool, an Enterprise seat doesn't change what agent work costs.
- Allowances: we assume typical developers fit inside a standard Claude or Codex seat and heavy users inside a premium one. If they don't, usage credits add to the seat cost.
- Copilot: two rows. One is completions and light chat, which costs the seats. The other is the same agent work run through Copilot, priced as AI credits at per-token rates after the pooled allowance. That assumes similar tokens on similar models, since each tool's agent uses tokens differently.
Scenario 1: mostly light users: what 50 developers cost
Most developers use AI for autocomplete, chat and the occasional agent task. A few run agents daily. 30 occasional, 17 typical and 3 heavy agent users. Their agent work comes to $5,805 a month at API rates, about $116 per developer.
| Option | Monthly cost | Per developer | What it assumes |
|---|---|---|---|
| Claude Code, Claude Team | $1,240 + any usage credits | $25/dev | 47 Standard seats, 3 Premium seats; usage stays inside the allowances |
| OpenAI Codex, ChatGPT Business | $1,240 + any credits | $25/dev | 47 Standard, 3 Premium seats; usage stays inside the limits |
| GitHub Copilot, completions and light chat | $1,010 | $20/dev | 47 Business, 3 Enterprise seats; completions don't use AI credits |
| GitHub Copilot, same agent work | $5,805 | $116/dev | The scenario's agent work billed as AI credits at per-token rates, after the pooled allowance |
| Claude Code on API billing | $5,805 | $116/dev | Every token at Anthropic API rates, no seats |
| Claude Enterprise, usage-based | $6,805 | $136/dev | $20 seat plus usage at API rates |
Verdict: if your team mostly uses autocomplete and chat, Copilot is the cheapest at $1,010 a month. Completions don't use credits, and the pooled credits cover light chat. Once the 17 typical developers run agents most days, Claude Code or Codex seats at $1,240 cost far less than running the same agent work through Copilot ($5,805). Pure API billing makes sense only if most of the team barely uses agents at all.
Scenario 2: a mixed team: what 50 developers cost
Most developers use an agent most days. A fifth run long agent sessions every day. 10 occasional, 30 typical and 10 heavy agent users. Their agent work comes to $12,350 a month at API rates, about $247 per developer.
| Option | Monthly cost | Per developer | What it assumes |
|---|---|---|---|
| Claude Code, Claude Team | $1,800 + any usage credits | $36/dev | 40 Standard seats, 10 Premium seats; usage stays inside the allowances |
| OpenAI Codex, ChatGPT Business | $1,800 + any credits | $36/dev | 40 Standard, 10 Premium seats; usage stays inside the limits |
| GitHub Copilot, completions and light chat | $1,150 | $23/dev | 40 Business, 10 Enterprise seats; completions don't use AI credits |
| GitHub Copilot, same agent work | $12,350 | $247/dev | The scenario's agent work billed as AI credits at per-token rates, after the pooled allowance |
| Claude Code on API billing | $12,350 | $247/dev | Every token at Anthropic API rates, no seats |
| Claude Enterprise, usage-based | $13,350 | $267/dev | $20 seat plus usage at API rates |
Verdict: Claude Code or Codex seats, with premium seats for the 10 heavy users, at $1,800 a month. That is about a seventh of the same work at API rates ($12,350), as long as allowances hold. Watch the heavy users: if they outgrow premium allowances, usage credits are billed at API rates on top. Moving them to Copilot wouldn't help, because Copilot charges per-token rates too once its pool is gone.
Scenario 3: heavy agent users: what 50 developers cost
Agents do much of the work. Most developers run long sessions all day, on large codebases. 5 occasional, 15 typical and 30 heavy agent users. Their agent work comes to $21,925 a month at API rates, about $439 per developer.
| Option | Monthly cost | Per developer | What it assumes |
|---|---|---|---|
| Claude Code, Claude Team | $3,400 + any usage credits | $68/dev | 20 Standard seats, 30 Premium seats; usage stays inside the allowances |
| OpenAI Codex, ChatGPT Business | $3,400 + any credits | $68/dev | 20 Standard, 30 Premium seats; usage stays inside the limits |
| GitHub Copilot, completions and light chat | $1,550 | $31/dev | 20 Business, 30 Enterprise seats; completions don't use AI credits |
| GitHub Copilot, same agent work | $21,925 | $439/dev | The scenario's agent work billed as AI credits at per-token rates, after the pooled allowance |
| Claude Code on API billing | $21,925 | $439/dev | Every token at Anthropic API rates, no seats |
| Claude Enterprise, usage-based | $22,925 | $459/dev | $20 seat plus usage at API rates |
Verdict: premium seats are where subscriptions save the most. $3,400 a month on Claude Code or Codex covers work worth $21,925 at API rates, if 30 heavy users stay inside premium allowances. Codex Premium has no 5-hour limit, but weekly limits can apply. Claude Premium has 5x Standard usage, with 5-hour and weekly windows. At this intensity, measure who outgrows their seat before you commit to one tool.
Codex vs Claude Code vs Copilot: all three scenarios side by side
| Option | mostly light users | a mixed team | heavy agent users |
|---|---|---|---|
| Claude Code, Team seats | $1,240 + credits | $1,800 + credits | $3,400 + credits |
| Codex, Business seats | $1,240 + credits | $1,800 + credits | $3,400 + credits |
| Copilot, completions and light chat | $1,010 | $1,150 | $1,550 |
| Copilot, same agent work | $5,805 | $12,350 | $21,925 |
| Claude Code, API billing | $5,805 | $12,350 | $21,925 |
| Claude Enterprise, usage-based | $6,805 | $13,350 | $22,925 |
Claude Code and Codex have the same seat prices, so between them the choice comes down to models, workflow and how far each seat's allowance stretches for your work. Run a two-week pilot of each with the same developers before you sign an annual contract. The calculator also compares Cursor, which has its own weekly pools and Premium, Super and Ultra seats.
What changes the answer for your team?
- The model. Opus-class models cost more per token than Sonnet-class models: Claude Opus 5.5 is twice Sonnet 5.5. They also use up seat allowances faster. See Opus or Sonnet in Claude Code.
- Autocomplete vs agents. Copilot's completions are unlimited and don't use credits. If that is most of your use, Copilot's seat price is close to the whole cost.
- Unpublished allowances. Anthropic and OpenAI don't publish seat allowances in dollars. The seat rows above are a floor, and usage credits are billed on top when developers outgrow them.
- Monthly billing. Claude Code and Codex seats cost 25% more month to month: $2,250 instead of $1,800 for the mixed team.
- Two tools, not one. Many teams keep Copilot for completions and add Claude Code or Codex for agent work. That adds seat costs, so check how many developers really use both.
How do you find out which scenario your team is in?
Measure, don't guess. For Claude Code, send its OpenTelemetry export to Ganivra. Every request is priced at API rates by developer and team, on any plan, so you can see who is occasional, typical or heavy, and what your mix would cost on each billing option. Roll it out with managed settings. Then read the numbers in Claude Code cost per developer. For Copilot, GitHub's AI usage report shows credits per user and model. If finance needs the result per team, see charging AI coding costs back to teams.
Codex vs Claude Code vs Copilot: frequently asked questions
Which is cheapest for a 50-developer team: Codex, Claude Code or Copilot?
On seat price, GitHub Copilot: $1,010 to $1,550 a month in our three scenarios, against $1,240 to $3,400 for Claude Code or Codex on annual billing. But Copilot charges agent work as AI credits at per-token rates once the pooled allowance runs out. When developers run agents most days, Claude Code and Codex seats usually cost less, because each seat includes far more agent usage than its price would buy at API rates.
How much does Claude Code cost for 50 developers?
On the Claude Team plan, $1,240 to $3,400 a month on annual billing in our scenarios, depending on how many developers need a Premium seat, plus any usage credits beyond the seat allowances. On API billing, the same work costs $5,805 to $21,925 a month at Anthropic's published averages.
How much does OpenAI Codex cost for a team?
Codex is included in ChatGPT Business. A Standard seat is $20 a month billed annually or $25 monthly, and a Premium seat, with more usage and no 5-hour limit, is $100 or $125. Seat types can be mixed. Usage beyond the limits continues on workspace credits, whose price depends on your plan or agreement. Codex can also run on an API key at standard API rates.
Are GitHub Copilot AI credits charged at API prices?
Each token is priced at the model's rate and converted to AI credits at $0.01 per credit. For Claude Sonnet 5.5 and GPT-6.1 Sol, GitHub's published per-token rates match Anthropic's and OpenAI's API list prices. Copilot Business includes 1,900 credits per user and Enterprise 3,900, pooled across the organization; code completions and next edit suggestions don't use credits.
Should heavy agent users get premium seats?
Usually, if they hit the standard seat's limits often. A Claude Premium seat has 5x the usage of a Standard seat, and a ChatGPT Business Premium seat has no 5-hour limit. On Copilot, an Enterprise seat adds $20 of credits to the pool for $20 more, so above the pool it doesn't make agent work cheaper.
Can I run the numbers for my own team?
Yes. The Ganivra AI coding tools cost calculator compares Claude Code, Cursor, Copilot and Codex seat costs for any number of everyday and heavy users, shows what the same team costs on API billing, and when self-hosting an open model pays off. Each scenario in this article links to the calculator with its inputs filled in.
Your team, your mix
Price your own team in the AI coding tools cost calculator.
Set how many everyday and heavy users you have. The calculator compares Claude Code, Cursor, Copilot and Codex seats, API billing, and self-hosting an open model. It's free, with no sign-up.