How Much Does ChatGPT for Business Cost? Plans, Per-Seat Prices, and Hidden Usage Fees (2026)

How much does ChatGPT for business cost is now a three-part question, because OpenAI's 2026 pricing has quietly split into three layers: published seats, a quote-only enterprise tier, and a metered usage layer that most pricing guides still ignore. The short answer: ChatGPT Business runs $20 per user per month billed annually ($25 monthly), Enterprise is quote-only with 2026 procurement reports converging around $45-$75 per seat and a reported ~150-seat minimum, and heavy usage on top of either plan is billed separately through token or credit pricing. This guide breaks down each layer with real numbers, where the metered fees bite, and the point at which a private, self-hosted AI stack becomes the cheaper option.

ChatGPT Plan Pricing at a Glance (2026)

• Free: $0, ad-supported in the US since February 2026, with tight limits on uploads and tools.
• Go: about $8/month - expanded limits for individuals.
• Plus: $20/month - full individual feature set; the reference point most comparisons use.
• Pro: $100/month for the entry tier, $200/month for maximum usage (power users, heavy Codex and agent work).
• Business (formerly Team): $20/user/month annual or $25 monthly, 2-seat minimum. OpenAI cut this price by $5/seat on April 2, 2026 - guides still quoting $25/$30 are out of date.
• Enterprise: no published price; contact-sales only, annual prepay, with a reported ~150-seat minimum.

A Premium Business seat with roughly five times the usage allowance costs $100 annually or $125 monthly. At five or more users, Business annual beats the equivalent number of Plus subscriptions - and adds SSO, SOC 2, an admin console, and a contractual no-training guarantee that consumer plans don't carry by default.

ChatGPT Enterprise: The Quote-Only Tier, Honestly Explained

OpenAI publishes no Enterprise price - the pricing page says only "contact sales." What 2026 procurement reports have converged on:

• $45-$75 per seat per month, around $60 typical, falling toward ~$40/seat for 5,000+ seat deals (Coworker AI's July 2026 procurement analysis, citing beam.cloud and teamazing reports).
• A reported ~150-seat minimum with annual prepayment - implying an entry point near $108,000 per year (150 × $60 × 12).
• What the premium buys: SCIM provisioning and RBAC, invoicing/PO payment, data residency (10 storage regions, Enterprise-only, free), a Compliance API for eDiscovery/DLP/SIEM, customer-managed encryption keys (added December 2025), custom SLAs, and a HIPAA BAA on request - none of which Business offers.

One caveat worth more than the price itself: the famous "$60 per seat" figure traces back to a single 2023 Reddit post by one buyer sharing their quote, and OpenAI has never confirmed it. Treat $60 as a budgeting anchor, not a list price - and get the seat count, minimum, renewal terms, and usage conditions in writing before comparing quotes.

The Hidden Third Layer: Usage and Token Pricing

This is where 2026 budgets actually break, and where most pricing articles stop. Seat fees cover included usage; heavy use gets metered:

• Credit-based Business/Enterprise billing: high-reasoning messages cost 10 credits per message on OpenAI's published rate card; GPT-6 Pro costs 50 credits per message.
• Enterprise token-based billing replaces message counts with raw token rates - published rates run from $0.10/$0.50 per million input/output tokens for the lightest Work model up to $10.00 input / $50.00 output for the flagship (GPT-6 Astra class).
• Codex (coding agent) usage is estimated by OpenAI at roughly $100-$200 per developer per month depending on model choice and concurrency.

OpenAI's own rate-card help pages document this shift from per-message to token-based enterprise billing, so if your team's usage is spiky - agent runs, long documents, code generation - budget a usage line item on top of the seat bill, and ask the sales team to model it against your last quarter's activity.

Self-Hosted AI vs ChatGPT: Where the Crossover Sits

For comparison shopping, the alternative is a private stack running open-weight models (Llama, Qwen, Mistral, gpt-oss) on your own hardware. The honest structure of that comparison:

• Cost curve: per-seat fees scale linearly with headcount - 50 employees on Business annual is $12,000 every year, forever. Self-hosted has a one-time hardware cost (practical entry tiers start near $2,000; 70B-class rigs run $4,000-$10,000; frontier-grade builds approach $40,000) plus electricity and maintenance, with no per-seat charge - the 200th user costs the same as the first.
• Control: self-hosted runs on your servers, can be air-gapped, and keeps prompts and documents entirely inside your perimeter - Enterprise data residency narrows the gap but the processing still happens in OpenAI-operated cloud.
• Model choice: any open model, switchable anytime, versus locked to OpenAI's models.
• The catch: self-hosted requires infrastructure skill or a deployment partner, and open models trail frontier GPT models on the hardest reasoning and coding tasks. Convenience still favors ChatGPT; ownership favors self-hosting.

The crossover is simple to compute: take your annualized seat bill, subtract one year of hardware-plus-power cost for a right-sized private stack, and look at the three-year total. Teams over roughly 50 daily AI users - or any team with strict data-control requirements - usually find the private stack wins on three-year cost; a five-person shop almost never does.

What Each Business Tier Actually Includes

Price is only half the comparison - the tiers differ in what they contractually guarantee:

• Business: admin-managed workspace, SAML SSO, SOC 2 Type II and ISO 27001, no-training-on-your-data by default (no toggle needed), business-app connectors, and usage caps per seat - with an agent-mode allowance that many teams exhaust mid-month, which is the most common trigger for an Enterprise conversation.
• Enterprise: everything above plus SCIM provisioning and role-based access control, invoicing or purchase-order payment (Business is card-only), data residency in 10 storage regions, a Compliance API covering eDiscovery, DLP and SIEM integration, customer-managed encryption keys, custom retention periods and SLAs, volume discounts, and a HIPAA BAA on request.
• What neither tier changes: the processing still happens on OpenAI-operated infrastructure. Enterprise narrows the compliance gap with contractual controls; it does not move the computation into your building. If your requirement is "no third party processes this data," neither tier satisfies it - that is the on-premise conversation.

How to Choose: A Decision Path

1. 1-3 users, general productivity: Plus at $20/month each - simplest and cheapest.
2. 4-100 users, standard business use: Business at $20/seat annual. SSO, SOC 2, and no-training-by-default justify the switch from personal Plus accounts immediately.
3. Power users hitting caps: mix in Premium Business seats ($100/$125) or a couple of Pro subscriptions rather than upgrading the whole team.
4. Regulated industries, health data (HIPAA), 150+ seats, or audit requirements: Enterprise quote - or evaluate a private deployment, because at $108K+/year the self-hosted alternative is genuinely competitive.
5. Agent- or Codex-heavy workflows: price the metered layer first; it can exceed the seat bill.

Three Procurement Mistakes to Avoid

Before you enter a card number or book an Enterprise demo, the recurring failure patterns:

• Buying seats before measuring usage. Rollouts purchased at headcount routinely show a meaningful share of seats going quiet within a quarter. Start with the daily-user count, not the payroll - the plan choice changes at roughly four regular users, not at the size of the company.
• Ignoring the metered layer until the invoice arrives. Agent runs and Codex usage are billed separately; OpenAI estimates $100-$200 per developer per month for heavy Codex work. Model your spikiest month, not your average one.
• Treating Enterprise quotes as fixed. Seat minimums, term length, and usage allowances all move in negotiation. The reported $60/seat anchor is a starting point; buyers who bring modeled usage data consistently land better terms than buyers who accept the first quote.

Frequently Asked Questions

How much does ChatGPT Business cost per user?

$20 per user per month billed annually or $25 billed monthly, with a 2-seat minimum, following OpenAI's April 2, 2026 price cut from $25/$30. A Premium seat with five times the usage runs $100 annual or $125 monthly.

How much does ChatGPT Enterprise cost?

OpenAI publishes no price. 2026 procurement reports cluster at $45-$75 per seat per month (~$60 typical) with a reported 150-seat minimum and annual prepay - an entry point near $108,000 a year. Treat every number as a negotiating anchor until your quote arrives.

Is ChatGPT Business safe for company data?

More than consumer plans: Business excludes your data from training by default, carries SOC 2 Type II and SAML SSO, and Enterprise adds data residency, customer-managed keys, and a Compliance API. But "not trained on" is not "never leaves our cloud" - the processing still happens in OpenAI-operated infrastructure, which is why regulated buyers consider private deployments.

What is the difference between ChatGPT Plus and Business?

Business is the same per-seat price as Plus when billed annually, but adds workspace administration, SSO, SOC 2 compliance, connectors to business apps, contractual no-training terms, and higher usage caps. For any team of four or more, Business is the correct tier.

Does ChatGPT Business include Codex and agents?

Included usage covers standard chat and tools; Codex and agent-heavy usage is metered separately - OpenAI estimates roughly $100-$200 per developer per month depending on model and concurrency. Budget the metered layer before rolling agents out to a whole team.

Choosing between seat-based AI and a private, self-hosted stack is a cost-and-control question, not a fashion question - contact ArcBeta Solutions for a straight answer on your team's numbers. For the full breakdown of private infrastructure costs, see our guide to on-premise AI server costs for small business.