AI cost per team in 2026: Claude Code and Copilot
Seats, usage, rollout: what Claude Code, GitHub Copilot and Microsoft 365 Copilot cost according to their price pages, with source and date, and how to control usage.
Sebastian LangOctober 5, 202614 min read
Anyone planning AI agents for a whole team first asks about the Claude Code costs or about Copilot licence costs for the business. The answer is on the vendors' price pages, but it is only part of the calculation. Besides the seat price, additional usage and rollout costs can arise.
This post collects the list prices of Anthropic, GitHub and Microsoft as they stood on the price pages on 5 October 2026, each with source and currency. It also explains why usage costs are harder to plan than seat costs, provides a worked example with disclosed assumptions, and describes ways to control costs. "Copilot" here means two things: GitHub Copilot for developer teams and Microsoft 365 Copilot for office work.
TL;DR
- An AI workstation involves three types of cost: seats, usage and rollout. With Claude Enterprise, usage at API prices comes on top of the seat; with GitHub Copilot, everything beyond the included credit pool is billed by consumption.
- List prices on 5 October 2026: Claude Team from 20 USD per seat per month (annual), GitHub Copilot Business 19 USD per seat per month, Microsoft 365 Copilot €26.00 per user per month (annual, excl. VAT).
- Gartner forecasts rising costs per agentic workflow even though tokens are getting cheaper. Anyone who fixes budgets, model choice and measurement before the rollout keeps the bill under control.
The three types of cost: seats, usage, rollout
Seats. This is the fixed monthly fee per person shown in the quote. Once plan, seat type and billing interval are fixed, the seat fee can be calculated from the number of people.
Usage. These are costs that rise with consumption. According to Anthropic, Claude Code bills by consumed API tokens, and the cost per developer varies widely with model choice, codebase size and way of working, for example when several instances run in parallel or automation is involved (Claude Code Docs). Claude Pro, Max and Team include a usage allowance in the seat price. Anyone who exceeds it can add usage credits at standard API prices (Claude Pricing). The current Enterprise plan works differently: it charges a seat fee plus all usage at API prices, with no allowance per seat. Only older, seat-based Enterprise plans have an allowance per seat (Claude Help Center). With GitHub Copilot, each seat brings an allowance of GitHub AI Credits, and extra consumption is billed per credit (GitHub Docs).
Rollout. The table prices do not reflect your internal effort for training, setup and pilot phase. Anthropic recommends starting with a small pilot group and measuring a baseline with the tracking tools before you roll out more widely (Claude Code Docs). What such a rollout looks like with configuration and governance is described in Rolling out Claude Code in your team. If you first want to clarify what the tool actually does: What is Claude Code?
Make it measurable
How to tell whether AI is working
What the vendors charge today
The following prices come from the vendors' price pages and documentation, retrieved on 5 October 2026. Anthropic and GitHub quote US dollars, the German Microsoft pages euros. Anthropic points out that the prices shown exclude taxes; Microsoft quotes prices excl. VAT.
| Product and plan | List price (as of 5 October 2026) | What is said about usage |
|---|---|---|
| Claude Pro | 20 USD per month, 17 USD per month with annual payment (200 USD upfront) | Claude Code included, usage limits apply |
| Claude Max | from 100 USD per month, monthly only | 5x or 20x the usage of Pro |
| Claude Team, Standard seat | 20 USD per seat per month annually, 25 USD monthly | for teams of 2 to 150 people, more usage than Pro, Claude Code and Cowork included |
| Claude Team, Premium seat | 100 USD per seat per month annually, 125 USD monthly | 5x the usage of a Standard seat, seat types can be mixed |
| Claude Enterprise | 20 USD per seat per month, annual billing, plus usage at API prices | costs scale with model and task, admins set spend limits |
| GitHub Copilot Business | 19 USD per assigned seat per month | 1,900 AI Credits per person per month in a shared pool, beyond that 0.01 USD per credit |
| GitHub Copilot Enterprise | 39 USD per assigned seat per month | 3,900 AI Credits per person per month in a shared pool, beyond that 0.01 USD per credit |
| Microsoft 365 Copilot | €26.00 per user per month with annual billing, €27.30 with monthly payment on an annual subscription | separate licence for an eligible Microsoft 365 plan required |
| Microsoft 365 Copilot Business | €18.20 per user per month with annual billing, promotional price €15.60 in the first year (promotion 1 July to 31 December 2026, conditions according to Microsoft) | up to 300 users, eligible Microsoft 365 plan required |
| Microsoft Agent 365 | €13.00 per user per month with annual billing (US page: 15 USD) | control plane for agents, also included in Microsoft 365 E7 |
Sources: Claude Pricing, GitHub Copilot Plans, Microsoft 365 Copilot, Microsoft 365 Copilot Business, Microsoft Agent 365.
Two details change the calculation. With GitHub Copilot, the AI Credits of all licences flow into a shared pool that resets on the first day of each calendar month; unused credits expire. Code completions consume no credits and remain unlimited in all paid plans (GitHub Docs: Billing, GitHub Docs: Plans). In Pro, Max, Team and seat-based Enterprise plans, Claude chat and Claude Code draw from the same allowance; in Team and seat-based Enterprise plans, a seat's allowance is also shared with Cowork (Claude Pricing, Claude Code Docs).
What a token costs
Where billing is by consumption (Claude Enterprise, usage credits, API), the model prices per million tokens (MTok) apply. According to the Anthropic price page of 5 October 2026:
| Model | Input per MTok | Output per MTok | Cache read per MTok |
|---|---|---|---|
| Claude Opus 5.5 | 4 USD | 20 USD | 0.20 USD |
| Claude Sonnet 5.5 | 2 USD | 10 USD | 0.20 USD |
| Claude Haiku 4.5 | 1 USD | 5 USD | 0.10 USD |
For both input and output, Opus 5.5 therefore costs twice as much as Sonnet 5.5. Which model your team uses by default is thus a budget decision, not a matter of taste.
Why usage costs are hard to plan
The seat price multiplies cleanly with team size. Usage does not, for four reasons.
Consumption depends on the way of working. Anyone trying to plan the Claude Code cost per month will find a reference point at Anthropic: for Claude Code, the vendor cites an average across enterprise deployments of around 13 USD per developer per active workday and 150 to 250 USD per developer per month; for 90 percent of users, costs stay below 30 USD per active day (Claude Code Docs). That is an average, not a price. According to the same documentation, agent teams consume about 7x as many tokens as standard sessions when the team members run in plan mode, and each subagent sends its own requests.
Gartner expects consumption to rise with new agent capabilities. Gartner forecasts that inference costs per agentic workflow will rise to more than five times their current level by 2028 (forecast of 17 August 2026). Gartner calls this the "inference paradox": each new generation of agent capabilities consumes more tokens per task, often on a more expensive model tier, and faster than prices per token fall (Gartner).
The range is wide. In another forecast, dated 24 June 2026, Gartner expects AI coding costs to exceed the average developer salary by 2028, driven by token consumption and by vendors moving to usage-based licences. According to Gartner, many vendors do not transparently disclose how token consumption is calculated and billed (Gartner). According to an analysis of Gartner Peer Insights reported by Computer Weekly, 23 percent of tech leaders spend 200 to 500 USD per developer per month on tokens for AI coding agents, and 6 percent of organisations pay more than 2,000 USD per developer per month (Computer Weekly).
Prices and rules keep changing. On 22 September 2026, Anthropic released Claude Opus 5.5 at 4 USD per million input tokens and 20 USD per million output tokens; Opus 5 remains on the price list at 5 and 25 USD. According to Anthropic, Opus 5.5 is 40 percent cheaper to run than Opus 5 on typical workloads (Anthropic). On 30 July 2026, three weeks after the launch of the GPT-5.6 family, OpenAI cut the price of GPT-5.6 Luna by 80 percent and of GPT-5.6 Terra by 20 percent (OpenAI). Billing rules move too: Anthropic had announced that from 15 June 2026 the Agent SDK, claude -p and third-party apps would no longer draw from the subscription allowance but from a separate monthly credit. On 15 June the change was paused; according to the Help Center, this usage still draws from the subscription allowance, and Anthropic says it will announce changes before they take effect (Anthropic Help Center).
For your planning this means: cheaper tokens do not automatically mean a smaller bill. Plan AI agent costs as a range and review them regularly.
Worked example for a team
The following calculation is an illustration, not an offer and not a forecast. It uses only the list prices above and these assumptions:
- The team consists of 10 developers, all of whom use a coding agent.
- The list prices of 5 October 2026 apply, with annual billing for Claude Team. Taxes, discounts and exchange rates are not taken into account.
- In the mixed variant, 3 people work intensively with agents every day and get Premium seats, the other 7 get Standard seats.
- For Claude Enterprise, we assume usage at the average cited by Anthropic of 150 to 250 USD per developer per month. Your team may be above or below that.
- Rollout costs (training, setup, time for the pilot) are not included.
| Variant | Calculation | Fixed amount per month | What can come on top |
|---|---|---|---|
| Claude Team, 10 Standard seats | 10 x 20 USD | 200 USD | Usage credits at API prices, if enabled |
| Claude Team, 3 Premium and 7 Standard seats | 3 x 100 USD + 7 x 20 USD | 440 USD | Usage credits at API prices, if enabled |
| Claude Team, 10 Premium seats | 10 x 100 USD | 1,000 USD | Usage credits at API prices, if enabled |
| Claude Enterprise | 10 x 20 USD seat, plus 10 x 150 to 250 USD usage (assumption) | 200 USD seat fee | 1,500 to 2,500 USD usage by assumption, billed at API prices |
| GitHub Copilot Business | 10 x 19 USD | 190 USD | 19,000 AI Credits per month included in the pool, beyond that 0.01 USD per credit |
The table shows three things. First, the fixed part is quickly calculated: the Claude Team price for 10 Standard seats comes to 2,400 USD a year. Second, the variable component is decisive: in the Enterprise variant, the assumed usage costs are several times the seat fee. Third, the variants are only partly comparable, because GitHub measures included usage in credits and Anthropic in usage limits per seat.
For an office team, Microsoft 365 Copilot is calculated separately: 10 users times €26.00 comes to €260.00 per month with annual billing, excl. VAT and in addition to the eligible Microsoft 365 licence.
How to control costs
Usage costs cannot be predicted exactly, but they can be capped and steered. Five levers, all using tools the vendors document themselves.
1. Set limits before you enable usage credits
In Claude Team and in older, seat-based Enterprise plans, the allowance per seat is the default ceiling. Anyone who needs to work beyond it needs usage credits, and for those you set spend limits at the level of the organisation, a group or individual members. The current Enterprise plan has no allowance per seat; costs are controlled through admin spend limits. The spend report in the org analytics shows estimated spend per person and model, with CSV export. Anyone billing via the API through the Console sets spend limits per workspace (Claude Code Docs).
With GitHub Copilot Business and Enterprise, admins set budgets for individuals, cost centres, organisations or the whole enterprise. If the policy for paid extra usage is switched off, Copilot pauses once the pool is used up until the next cycle (GitHub Docs). For configured budgets there are alerts at 75, 90 and 100 percent (GitHub).
2. Match the model to the task
Anthropic writes that Sonnet handles most coding tasks well and costs less than Opus; Opus should be reserved for complex architecture decisions and multi-step reasoning, and for simple subagent tasks the documentation recommends Haiku. According to Anthropic, unexpectedly high costs under API billing usually trace back to long sessions that are never cleared or to Opus as the default model (Claude Code Docs). Gartner likewise recommends giving routine tasks to smaller models (Gartner).
3. Keep the context small
Every request sends the conversation so far along with it. The Claude Code documentation therefore recommends restarting with /clear between unrelated tasks, keeping CLAUDE.md under 200 lines and moving specialised instructions into skills that load only when needed. Keep agent teams small and shut them down when their work is complete (Claude Code Docs). How to separate CLAUDE.md and skills cleanly is shown in the three-layer architecture for coding agents.
4. Classify work and measure consumption
Gartner advises sorting work into three classes: developer-led, developer with agent and agent-led. Additional measures include monitored token thresholds and reviewing high-consumption workflows in sprint retrospectives. Analyst Nitish Tyagi warns: without a governed operating model for software development, costs can rise faster than productivity gains (Gartner).
Costs alone, however, say nothing about benefit. There is a free PDF on how to measure AI adoption. The background is provided by Measuring AI productivity honestly.
5. Train the team
Training can help the team choose models, maintain context hygiene and measure consumption. Our training costs €599 per licence, one-off, and includes the courses on Cowork and Claude Code (pricing). What training costs in general and which funding may apply is covered in AI training for employees: cost and funding.
If you want to go through the numbers for your team together: talk to us for 30 minutes.
FAQ
What does Claude Code cost per month?
Claude Code is included in all paid Claude plans. Pro costs 20 USD per month (17 USD with annual payment), Max from 100 USD, a Team Standard seat 20 USD and a Premium seat 100 USD per month with annual billing (as of 5 October 2026, excl. taxes). Under consumption-based billing, Anthropic cites an average from enterprise deployments of 150 to 250 USD per developer per month.
Is Claude Code included in the Pro plan?
Yes. According to Anthropic, Claude Code is included in all paid plans, including Pro. It is not included in the free plan.
Do Claude Code and Claude chat share the same allowance?
Yes, in Pro, Max, Team and seat-based Enterprise plans: there, Claude Code shares the plan's usage limits, and work in the terminal and chats draw from a common pool. The limits renew in a rolling five-hour window, and there are also weekly limits. In Team and seat-based Enterprise plans, Cowork is also part of this allowance per seat. The current usage-based Enterprise plan, by contrast, has no allowance per seat: all usage there is billed at API prices.
Why is the Claude Code bill higher than the plan price?
With Pro, Max and Team, the plan price covers only the included allowance. In the current Enterprise plan, usage at API prices comes on top of the seat fee. Anyone who adds usage credits pays standard API prices beyond that. Under API billing, Anthropic traces unexpectedly high costs mostly to long sessions that are never cleared or to Opus as the default model; subagents and agent teams that send their own requests add to this.
Is Claude Code cheaper than GitHub Copilot?
At the seat level, the list prices are close: GitHub Copilot Business costs 19 USD per seat per month, a Claude Team Standard seat 20 USD with annual and 25 USD with monthly billing. The two are only partly comparable, because GitHub measures included usage in AI Credits and Anthropic in usage limits per seat. Only measurements from a pilot with your team and your tasks can show which option is cheaper.
How do I avoid unexpected Claude Code bills?
Set spend limits before you enable usage credits: in Team and seat-based Enterprise plans for the organisation, groups or individual members, in the current Enterprise plan via the admins' spend limits, in the Console per workspace. Have your team restart with /clear between unrelated tasks and use Sonnet instead of Opus for most coding tasks. /usage shows your own consumption, and the spend report in the org analytics shows spend per person.
Sources
- Anthropic, Claude Pricing (retrieved 5 October 2026): claude.com/pricing
- Anthropic, Claude Code Docs, Manage costs effectively (retrieved 5 October 2026): code.claude.com/docs/en/costs
- Anthropic, Claude Opus 5.5 (22 September 2026): anthropic.com/claude-opus-5-5
- Anthropic Help Center, Use Claude Code with your Team or Enterprise plan (retrieved 5 October 2026): support.claude.com
- Anthropic Help Center, Use the Claude Agent SDK with your Claude plan (updated 16 June 2026): support.claude.com
- GitHub Docs, Plans for GitHub Copilot (retrieved 5 October 2026): docs.github.com
- GitHub Docs, Billing for organizations and enterprises (retrieved 5 October 2026): docs.github.com
- GitHub, Copilot Plans and pricing (retrieved 5 October 2026): github.com/features/copilot/plans
- Microsoft, Microsoft 365 Copilot (retrieved 5 October 2026): microsoft.com/de-de
- Microsoft, Microsoft 365 Copilot Business (retrieved 5 October 2026): microsoft.com/de-de
- Microsoft, Microsoft Agent 365 (retrieved 5 October 2026): microsoft.com/de-de
- Microsoft Security Blog, Microsoft Agent 365 now generally available (1 May 2026): microsoft.com
- Gartner, press release of 17 August 2026 on inference costs per agentic workflow (forecast): gartner.com
- Gartner, press release of 24 June 2026 on AI coding costs (forecast): gartner.com
- Computer Weekly, Gartner: AI coding agents will cost more than real developers (24 June 2026): computerweekly.com
- OpenAI, API Pricing: developers.openai.com
Make it measurable
How to tell whether AI is working
The KPI framework shows which metrics make AI adoption and productivity measurable. Free as a PDF by email.
About the author
Co-Founder · Business & Content Lead
Co-Founder of Sentient Dynamics. 15+ years of business strategy (incl. SAP), MBA. Writes about EU AI Act compliance, ROI measurement and how Mittelstand CTOs actually adopt agentic AI.
