
Photo by Archives Branch, USMC History Division, CC BY 2.0, via Wikimedia Commons
Small teams get billed by AI vendors in four different ways at once: a per-seat subscription (a fixed price for each named person), usage limits bundled into that seat, credits that unlock usage beyond those limits, and pay-per-token API billing for anything you build yourself. The cheapest setup for a five-person team is usually a base-tier seat for everyone, one or two higher-tier seats for the heaviest users, and no API spend at all until you have a specific automation worth paying metered rates for. Annual billing, credit pools, batch discounts, and prompt caching are all just ways of moving cost between those four buckets.
Prices change often and vary by region, so treat every number here as a documented snapshot from vendor documentation, not a quote.
Seat-based team plans: ChatGPT Business and Claude Team
Both major chat-first vendors sell team plans built on two seat tiers at identical list prices.
ChatGPT Business (OpenAI) prices Standard seats at $25 per user per month billed monthly, or $20 billed annually, and Premium seats at $125 monthly / $100 annually. Premium buys a higher usage allotment for ChatGPT Work and Codex and removes the five-hour usage limit. Workspaces usually require two paid seats. OpenAI documents all of this in its seat and billing management guide.
Claude Team (Anthropic) prices Standard seats at $20 per seat monthly on an annual plan ($25 monthly) and Premium at $100 / $125, described as 5x more usage than Standard. Team targets organizations of roughly 2 to 150 people; Anthropic's Claude pricing page lists it alongside individual Free, Pro ($17 monthly on annual billing, $20 monthly), Max ($100), and Ultra ($200) plans.
The seat mechanics matter more than the headline rate:
- You pay for seats, not activity. OpenAI charges for every Standard and Premium seat in the workspace at the start of each cycle "whether or not they've been assigned to individual members." An unassigned seat is pure waste.
- Seats are reusable, but reductions lag. Removing a member frees the seat for reassignment but does not cut your paid count on either platform. OpenAI requires a scheduled reduction effective at the next billing cycle; Anthropic states removing members produces no credit or refund.
- One seat, one human. Anthropic is explicit that each Team seat belongs to a named individual, credential sharing and routing work through someone else's remaining usage are prohibited, and reassignment is for real team changes rather than rotating one seat among several active users.
- Tier upgrades beat new purchases. If an unused Premium seat exists in your allocation, both vendors let you reassign tier instead of buying more capacity.
- Self-serve means card-only. OpenAI's self-serve Business workspaces accept credit and debit cards (and, where available, workspace wallet funds), but not invoicing, bank transfer, wire, ACH, purchase orders, or net terms; support cannot convert a self-serve workspace to invoice billing. Claude Team accepts credit, debit, or prepaid cards, not ACH. Invoicing and net terms appear only on contracted enterprise plans from both vendors — a real constraint if your finance process requires a purchase order.
Mid-cycle changes are prorated immediately. OpenAI's published example: a workspace with 10 Standard and 5 Premium seats ($875/month) that adds 3 Standard and 1 Premium on day 10 is charged $133.33 at once, then $1,075/month next cycle. Anthropic's example charges $3,616.44 immediately for upgrading five Standard seats to Premium with 275 days left on an annual term. Budget for hires and role changes, not just headcount at renewal.
Credits: three different things vendors call credits
1. Included usage limits. A seat includes a rate-limited allowance, not unlimited work. ChatGPT's pricing documentation publishes estimates ranging from 5–45 messages per five-hour period on GPT-6 Astra to 350–3,000 on GPT-6 Luna, and warns that "tasks that look similar can consume different amounts of your allowance" because model choice, context, reasoning, tool use, and caching all count. Anthropic applies 5-hour and weekly limits across Free, Pro, Max, Ultra, and Team.
2. Overage credits. OpenAI's flexible pricing lets Business workspaces buy shared credit pools: users have per-seat limits, and exceeding one draws from the workspace pool. With no credits available the feature is blocked until an owner adds more, and owners can enable automatic recharge. ChatGPT quotes credit rates per million input, cached input, and output tokens — 50 / 2.5 / 250 credits per 1M for GPT-6.1 Sol — and notes a typical GPT-5.6 Sol task may use 2–15 credits. Speed modes cost more: Fast consumes 2.5x included limits (2x on purchased credits), Ultrafast 8x / 6x.
Anthropic's equivalent is usage credits on Team and seat-based Enterprise plans: after a member hits the seat limit, further usage bills at standard API rates, with owners pre-purchasing a balance, setting an organization-wide monthly spend limit, and optionally enabling auto-reload. Per-member spend limits are not available to every Team organization. On usage-based Enterprise plans there is no allowance to run out of — everything bills at API rates from the first token.
3. Subscription-bundled API credits. Both vendors hand you API money as a plan benefit. ChatGPT Pro and Business subscriptions link to an OpenAI API organization for $20 per paid Standard seat or $100 per Premium seat, pooled and capped at $500 per workspace per month ($6,000 per annual cycle). Claude Max ($100/month), Ultra ($200/month), and Team (same per-seat rates, capped at $500/month) work the same way; Anthropic's example is three Standard plus two Premium seats = $260/month.
The fine print is nearly identical: the plan must be at least seven days old, credits expire at the end of the billing cycle with no rollover, you must own both the plan and the receiving API organization, and these API credits do not raise your ChatGPT or Codex limits.
API token pricing: the bucket that scales with your product
If your team only chats and runs agent tasks inside vendor apps, skip this. The moment you build a workflow — triage bot, document summarizer, internal classifier — you move to metered billing. Published standard-speed, short-context rates per 1M tokens (USD):
| Vendor / model | Input | Output | Cached input | Batch input / output |
|---|---|---|---|---|
OpenAI gpt-6-luna |
$0.10 | $0.50 | $0.01 | $0.05 / $0.25 |
OpenAI gpt-6.1-sol |
$2.00 | $10.00 | $0.10 | $1.00 / $5.00 |
OpenAI gpt-6-astra |
$10.00 | $50.00 | $1.00 | $5.00 / $25.00 |
| Claude Haiku 4.5 | $1 | $5 | $0.10 | $0.50 / $2.50 |
| Claude Sonnet 5.5 | $2 | $10 | $0.20 | $1 / $5 |
| Claude Opus 5.5 | $4 | $20 | $0.20 | $2 / $10 |
| Gemini 4 Carbon | $2, then $4 from Dec 1, 2026 | $10, then $20 | $0.10, then $0.20 | Batch API: 50% off |
Sources: OpenAI API pricing, Anthropic model pricing, Gemini Developer API pricing.
Three levers move a metered bill more than the model name does:
- Batch processing. Both OpenAI and Anthropic discount batched asynchronous work by 50% on input and output. Overnight document runs, embeddings, and evaluation sweeps belong in batch.
- Prompt caching. Repeated system prompts, style guides, and long thread history re-read at a fraction of input price — 0.1x base input on many Claude models, roughly 10% of standard input on OpenAI's mid-tier models. Cache writes carry a premium (1.25x for a 5-minute Claude cache, 2x for 1-hour), so caching only pays when content is genuinely reused.
- Speed and context tiers. OpenAI charges 2x for Fast endpoints and up to 6x for Ultrafast, prices long-context requests over 272K input tokens higher, and adds a 10% uplift on regional data-residency endpoints for models released on or after March 5, 2026. Google structures API access into tiers with hard monthly caps — $250 at Tier 1, $2,000 at Tier 2, $20,000–$100,000 at Tier 3 — a blunt but useful guardrail for teams with no FinOps function.
How to price a five-person team
- Separate the two jobs. Count people who need AI in daily work; separately list the automations you want to run. Seats buy the first, tokens buy the second. See our guide to choosing AI tools for a small business before you commit to either.
- Buy the minimum viable seat count. ChatGPT Business generally needs two paid seats; Claude Team starts at two. Base tier for everyone, nothing else, for the first cycle.
- Upgrade individuals, not plans. After 30–60 days, check who actually hits limits — ChatGPT's usage dashboard reports current limits and reset times, and Anthropic publishes Team usage analytics. Moving two heavy users to Premium is the difference between $25 and $100 per person per month.
- Decide whether overage is acceptable. Overage is metered convenience: Claude Team bills it at API rates, ChatGPT Business draws it from a purchased pool. Set the organization spend limit before enabling auto-recharge.
- Claim the API credits you already paid for. Both vendors grant up to $500/month based on seat count, with no card required on the API account. Link once at the start of a cycle; an unclaimed grant expires unused.
- Batch the non-urgent volume and cache shared context. Two 50%-class discounts on the same workload is the cheapest optimization available, and it requires no negotiation.
- Anchor your renewal date. Reductions take effect only at the next cycle on both platforms, and annual plans lock added seats until the term ends. If headcount may shrink, stay monthly until you are certain.
Rough list price for five seats: all Standard is $125/month billed monthly or $100 annually on either platform; moving two people to Premium takes it to $325 monthly or $260 annually. Taxes, mid-cycle proration, overage credits, and any API spend sit on top — and neither vendor's self-serve tier includes tax in the published figures.
FAQ
Is annual billing actually cheaper? Yes on list price — ChatGPT Business Standard drops from $25 to $20 per seat and Premium from $125 to $100, and Claude Team moves the same way. Anthropic also lists discounted Team plans for nonprofits and scientists at the same annual-equivalent per-seat rates. The trade is commitment: OpenAI notes added seats on annual plans stay committed through the current term, and reductions apply only at the next annual cycle. Unstable headcount is a good reason to pay the higher monthly rate.
Can we share one seat among several people? No. Claude Team seats are assigned to one named individual, and sharing credentials or routing work through another user's remaining usage is prohibited. ChatGPT Business lets owners reassign unused seats freely, but a seat is still one account for one person.
What is the difference between plan credits and API credits? Plan credits buy usage inside the subscription product when you exceed seat limits. API credits are a subscription benefit applied to a linked API organization for building with the Responses, Agents, Images, or Realtime APIs on OpenAI, or the Claude API. ChatGPT's bundled API credits do not raise ChatGPT or Codex limits, and Claude's monthly API credits are never charged to your Claude plan.
Do unused credits roll over? Generally no. OpenAI states unused subscription-bundled API credits expire at the end of the billing cycle; annual Business workspaces get a full year's grant in one issuance that still expires with the cycle. Anthropic's Team credits appear monthly with a stated expiry date.
How do we stop a runaway API bill? Use the guardrails already shipped: Gemini's tier spend caps pause service once an account hits its monthly limit; Claude Team usage credits support an organization-wide monthly spend limit; ChatGPT Business exposes spend controls and usage analytics, and credit-billed enterprise workspaces can set an overage limit where 0 blocks new credit-based requests. If limits keep biting, our piece on free AI tool tiers and their limits explains where the no-cost options stop.
Should a small team use an API at all? Only when a workflow must run without a human in a chat window — document pipelines, classification, triage, embedded agents. For everything else, the seat plan is cheaper and easier to cancel.
Sources: - OpenAI Help Center — Managing billing and seats in ChatGPT Business; Flexible pricing for the Enterprise, Edu, and Business plans - OpenAI Developers — API credits for ChatGPT subscribers; API pricing; ChatGPT pricing docs - Anthropic — Claude plans & pricing; model pricing documentation; Team plan billing FAQs - Claude Help Center — Purchase and manage seats on Team plans; Monthly API credits for Max, Ultra, and Team plans - Google AI Studio — Gemini Developer API pricing; billing tiers