How to negotiate an OpenAI enterprise contract. Seats, tokens, commit and renewal, priced on real use.

Price the OpenAI enterprise contract on the seats people use, the tokens your applications consume and a commit that ramps with adoption. Start the renewal at least 9 months out, with a costed alternative ready.

OpenAI no longer sells enterprises a handful of individual subscriptions. It sells a structured commercial package across ChatGPT Enterprise, ChatGPT Team and API consumption, and the opening quote usually prices coverage for your entire generative AI deployment, sized as if you were at the top of its customer scale.

The contract you want matches what you run: the seats people use, the tokens your applications consume, a commitment you will burn down, and a renewal plan you control. In our negotiations, pricing the deal that way saved 20 to 35 percent against OpenAI's opening enterprise quote.

What does an OpenAI enterprise contract cover?

An OpenAI enterprise contract covers up to four product lines and five commercial dimensions. The product lines are ChatGPT Enterprise, ChatGPT Team, the OpenAI API and any custom work. The dimensions are what you negotiate across all of them:

1. Product scope. Which of the four lines go into the contract, sized to what you have deployed.
2. Seats. The ChatGPT seat count, measured against the seats people use over the contracted term.
3. API consumption. Input tokens, output tokens and the wider consumption pattern, broken down by model.
4. Commit. The dollar commitment over the term, compared with your actual OpenAI spend.
5. Renewal plan. How scope, commit and price carry into the next term.

The five dimensions feed each other. A seat count that is too high inflates the commit. A commit sized on headcount ignores the API, where growth usually sits, and a renewal that starts late leaves no time to fix either.

The OpenAI deal also runs alongside your other GenAI renewals, consumption agreements and support contracts, so treat it as one event in that calendar. The AI platform contract negotiation playbook sets out the sequence across vendors.

What sits in each of the four product lines?

• ChatGPT Enterprise. The seat product for large deployments, with single sign on, fine grained access controls and audit logs. Enterprise workspaces also draw on a shared credit pool for Deep Research, Thinking models, image generation, Advanced Voice and Codex, with credit allocation and expiry set in your Order Form.
• ChatGPT Team, now ChatGPT Business. The self serve plan for smaller groups and lighter users, renamed on August 29, 2025. A Standard seat lists at $20 per user per month billed annually or $25 billed monthly, a Premium seat at $100 or $125, with a 2 seat minimum.
• OpenAI API. Metered consumption across the model lineup. The contracts we negotiated in 2024 and 2025 priced GPT-4o, GPT-4 Turbo, GPT-3.5 Turbo and the o1 reasoning model. The current flagship family is GPT-5.6, sold as Sol, Terra and Luna.
• Custom work. Custom integrations, dedicated capacity, fine tuning and similar arrangements, usually reserved for the largest customers. OpenAI sells Scale Tier and Reserved Capacity only through its sales team.

Price each line against what you have deployed today and what you have a funded plan to deploy. Let actual use define the scope that goes into the contract, line by line, before any discount is discussed.

How do the OpenAI commercial lines compare?

Each of the four lines meters differently, so each needs its own negotiating approach.

How many ChatGPT Enterprise seats should you buy?

Buy Enterprise seats for the people who use the product, and plan for everyone else separately. OpenAI meters ChatGPT Enterprise on seat count, so every seat costs the per seat subscription for the whole term whether or not anyone logs in. A typical seat base splits into four groups:

• Active seats. Employees who use ChatGPT Enterprise as part of their work.
• Dormant seats. Leavers whose accounts were never removed, people who tried the product once, and other stranded or low use accounts.
• Contractor seats. Contractor and temporary worker accounts in your workspace, often created outside the HR joiner and leaver process.
• Specialty seats. Special seat arrangements negotiated by the largest customers.

Dormant and low use seats made up 15 to 35 percent of the ChatGPT Enterprise base in our engagements. Remove or reclassify them before OpenAI prices the renewal. Then negotiate the per seat price against OpenAI's opening seat position, never against your worst case headcount. The GenAI knowledge hub has more on seat sizing.

A worked seat example

Say you hold 5,000 ChatGPT Enterprise seats. OpenAI does not publish Enterprise pricing, so this example assumes a contracted rate of $50 per seat per month purely for the arithmetic.

Even with a growth buffer, the right sized contract costs $624,000 a year less than the current base. A 10 percent discount on all 5,000 seats would save $300,000. Removing the dormant seats saves more, and whatever discount you win still applies to the seats that remain.

How do you check your own seat position?

• Workspace analytics. The ChatGPT Enterprise analytics dashboard shows active users and per user activity, and admins can export a User report for a custom date range of up to 12 months. Pull at least the last three months, since a single month flatters seats that were used once for a pilot.
• Your identity provider. If you provision through SCIM, compare the ChatGPT group with the HR leaver list. Accounts for people who have left are the easiest seats to remove.
• Contractor records. Match contractor accounts against active purchase orders and end dates.
• Credit consumption. Check which teams draw on the shared credit pool. Heavy users justify an Enterprise seat, while occasional users may fit ChatGPT Business or a smaller pool.

How is the OpenAI API priced and metered?

OpenAI meters the API per million tokens, with separate rates for input, cached input and output. Output tokens carry a premium over input, and at scale API consumption can rival or exceed what you spend on seats. The bill has four parts:

1. Input tokens. Everything sent to the model: prompts, instructions, retrieved documents and conversation history.
2. Output tokens. Everything the model produces, including reasoning tokens on reasoning models. Output usually drives the largest share of API cost.
3. Cached tokens. Input served from the prompt cache at a steep discount, which produces large savings on applications that resend long context.
4. Custom API arrangements. Batch processing, dedicated capacity, fine tuning and other arrangements for the largest customers.

On the current API pricing, output costs six times input on all three models. Cached input costs one tenth of the fresh rate, a 90 percent discount, while cache writes on GPT-5.6 cost 1.25 times the input rate. A cached prefix must be at least 1,024 tokens and stays eligible for 30 minutes after its last use.

The Batch API takes 50 percent off inputs and outputs for work that can wait up to 24 hours. Flex processing trades slower responses for lower prices on work outside production. For negotiated rates to compare against, see our OpenAI enterprise pricing benchmarks.

How do you check your own API consumption?

Pull three months of your own consumption before you model anything. The Usage API and Costs API, called with an organization admin key, return token counts and spend grouped by model and project, including cached input tokens. That gives you the input, output and cache split you need for a model like the worked example below.

A worked API example

Say one application consumes 4 billion input tokens and 1 billion output tokens a month on GPT-5.6 Terra. Compare a negotiated discount with two changes you can make yourself: caching half the input, then routing 40 percent of traffic to Luna for tasks your own testing shows Luna handles well.

Scenario C splits the 4 billion input tokens three ways, per million: 2 billion read from cache at $0.20, 200 million written to the cache at $2.50 (1.25 times input), and 1.8 billion fresh at $2.00. Output is 60 percent of the bill in scenario A, in line with what we see in live deployments.

Caching and routing cut the annual bill by 47 percent, about three times what the 15 percent discount delivers. Do that work first, then negotiate API rates and the commit shape against OpenAI's opening API position, using the smaller and better understood volume.

What happens when the models named in your contract are retired?

Most models named in the 2024 and 2025 contracts we reviewed now have shutdown dates. GPT-3.5 Turbo, GPT-4 Turbo and the May 2024 GPT-4o snapshot shut down on October 23, 2026, and o1 on December 11, 2026, with GPT-5.6 models as replacements. The plain GPT-4o alias has no date yet.

OpenAI's policy gives at least 6 months notice for generally available models, at least 3 months for specialized variants and as little as 2 weeks for preview models. A rate card tied to named models loses value each time one is retired. Tie discounts to total spend or to the successor model instead.

How should you shape the OpenAI commit?

Size the commit to the OpenAI spend you are confident of, and shape it as a ramp that follows adoption. OpenAI will propose a commitment sized to broad coverage. Under its standard terms, a minimum commitment cannot be cancelled except as required by law or expressly permitted. Commit shapes fall into four patterns:

1. Under commit. You commit below expected spend and pay a premium on overage. This is often the cleanest option when consumption is volatile.
2. At commit. You commit roughly in line with actual OpenAI spend.
3. Over commit. You commit above actual spend, which builds structural overspend into the contract.
4. Custom commit. Negotiated structures for the largest customers, often with phased ramps, true ups or burn down clauses.

A worked commit example

Say your plan shows API and credit spend of $600,000, $1,000,000 and $1,400,000 over three years, a total of $3,000,000. The table assumes unused commitment does not roll into the next year.

The flat shape commits $3,600,000 in total, so OpenAI will usually price it at a better rate. In the on plan case, that rate would have to recover $800,000 of unused commitment, more than a quarter of the $3,000,000 you actually spend. The reduction right is worth more than either shape when adoption stalls.

Why we advise against a large multi year commit

The usual advice is to lock a large multi year OpenAI commit to secure the best rate. We disagree for most enterprises. In most negotiations we ran, consumption was too volatile to commit confidently, and a flat over commit produced overspend that the rate never recovered. A phased ramp tied to actual adoption beat the big upfront commit.

Before you sign, read OpenAI's enterprise privacy terms and Microsoft's licensing terms for Azure OpenAI, which sells the same models. Our Azure OpenAI and direct OpenAI comparison sets the two routes side by side. Then shape the commit as a ramp, meter the API by model mix, and keep a costed Anthropic or Google alternative in play.

What have we seen in recent OpenAI enterprise negotiations?

Across roughly 20 to 30 OpenAI enterprise negotiations in 2024 and 2025, OpenAI opened broad and buyers were asked to pay for coverage they did not use. The customers who priced against actual seats and API consumption beat the opening quote, and OpenAI's business terms rewarded those who took the commit apart line by line.

Our engagement file for 2024 to 2025 holds 25 of those negotiations. The patterns that recur:

• Seats. Dormant or low use seats made up 28 percent of the ChatGPT Enterprise base across the file.
• Tokens. Output tokens drove 60 percent or more of API spend in most deployments.
• Savings. The median saving was 27 percent against OpenAI's opening enterprise quote.

OpenAI opens on broad coverage. You get a better contract by pricing the seats people use and the tokens your applications spend.

Where does the cost exposure sit in an OpenAI contract?

Cost exposure sits in four places, and each grows over the term unless the contract limits it. Our OpenAI contract risk review service works through all four alongside the data and IP terms.

1. Seat escalation. Seat growth during the term, often driven by wider rollouts that outrun the original sizing.
2. API consumption escalation. As OpenAI is built into your products and workflows, API consumption typically scales well above the contracted commit.
3. Commit drift. A gap between the contracted commit and actual OpenAI spend, in either direction, that produces avoidable cost.
4. Renewal escalation. OpenAI has a pattern of substantial increases at renewal for customers who have not run a structured renewal with enough lead time.

OpenAI's standard business terms, effective January 1, 2026, shape all four. The clauses that matter most for cost:

• Price changes. Changes on the public pricing page take effect 14 days after posting.
• Renewal notice. Notice of non renewal or scope reduction is due at least 30 days before the next renewal term.
• Liability. Each party's liability is capped at what you paid OpenAI in the prior 12 months, with carve outs for gross negligence, willful misconduct, indemnities and payment obligations.
• Termination. There is no general right to terminate for convenience. Either party can end the agreement for a material breach left uncured for 30 days, or for insolvency.
• Service reduction exit. If OpenAI notifies you of an update that materially reduces the service, you have 5 business days to elect termination on 30 days written notice.
• Commitment on exit. On any termination other than yours for OpenAI's breach, unpaid minimum commitment becomes due at once.

Contract wording to ask for

• Price hold. Fixed seat, credit and API rates for the term, so the 14 day pricing page clause cannot raise them. Our price hold clause language has sample wording.
• Renewal uplift cap. A stated maximum increase on seats and rates at renewal, which answers the renewal escalation risk directly. See our uplift cap redline.
• Successor model pricing. Your discount carries to the replacement model when a named model is retired.
• Commit reduction right. The right to reduce the next year's commit to trailing spend at each anniversary.
• Seat reduction and reassignment. The right to cut seats by a set percentage at each anniversary and to reassign seats from leavers.
• Credit pool terms. Rollover of unused credits and an overage limit set by you. Admins can set the overage limit to 0, which blocks credit use once the committed pool is spent.
• Early renewal proposal. OpenAI delivers its renewal quote at least 6 months before term end, well ahead of the 30 day notice deadline.

How should you prepare the OpenAI renewal?

Start at least 9 months before the renewal date and negotiate four areas separately. Take a multi year term only when the price protection is durable enough to justify the lock in.

• Renewal scope. The product and entitlement mix carried into the next term.
• Renewal commit. The dollar commitment for the next term, including any ramp or true up structure.
• Renewal term. The length of the renewal and the price protection that comes with it.
• Special arrangements. Custom renewal terms negotiated by the largest customers.

Keep a credible alternative in play across Anthropic Claude, Google Cloud Vertex AI and the wider GenAI field, costed on your own workloads. The Anthropic Claude enterprise licensing guide 2026 and our Vertex AI and Gemini negotiation guide give the competitive view.

Sequence the OpenAI renewal inside your wider GenAI renewal calendar, so OpenAI cannot negotiate with you in isolation while other AI contracts are still open. The AI contract renewal strategy enterprise playbook covers that calendar.

A renewal timeline

What will the OpenAI account team say, and how should you answer?

• "A larger commit gets you a better rate." Ask for rates at three commit levels, each with a reduction right, and compare them against unused commitment as in the worked example.
• "Enterprise wide coverage keeps governance simple." Reply that you will license measured use and add seats at the contracted rate as adoption grows.
• "The new model is more capable, so it costs more per token." Ask to price it on cost per completed task in your own testing, with your discount carried forward.
• "This pricing expires at the end of the quarter." Say your timeline follows your renewal date and your evaluation of alternatives.

How does the approach change with company size?

A 500 person company may find ChatGPT Business at list price covers its seat needs, with API spend on a pay as you go account. The decision there is mostly which plan fits. OpenAI points customers who need Zero Data Retention, a BAA, invoicing or purchase orders to its contracted offerings.

A 20,000 person enterprise faces every dimension at once: Enterprise seats, a credit pool, an API commit and possibly dedicated capacity. At that size the seat cleanup and the commit shape carry most of the value, and the Azure route becomes a real alternative for API consumption.

How does Redress work on OpenAI contracts?

We work only for buyers and take no commission from OpenAI or any reseller. The work comes in five forms:

• OpenAI scoping. A six week engagement that scopes the contract, measures your actual deployment and identifies what to change at the next renewal, run through our GenAI vendor services practice.
• OpenAI negotiation. We handle product scope, seats, API consumption, commit and the renewal itself, following the approach in the OpenAI enterprise procurement negotiation playbook.
• OpenAI contract risk review. A review of data terms, IP terms, indemnification and the wider contract risk.
• Vendor Shield. An always on advisory subscription that covers OpenAI alongside your other software vendors. See Vendor Shield.
• Spend assessment. The software spend assessment sizes the OpenAI conversation against your actual deployment.

What to do next

Run the OpenAI negotiation as a structured sequence with lead time.

1. Pull the deployment data. Actual seats, active use, token volume by model and your current commit.
2. Right size the seats. Reclassify or remove dormant and low use accounts before OpenAI quotes.
3. Model the API. Measure the output token share and test the savings from model mix and prompt caching.
4. Shape the commit as a ramp. Tie the commitment to measured adoption and include an annual reduction right.
5. Stand up an alternative. Cost Anthropic and Google on your own workloads so OpenAI knows the comparison exists.
6. Negotiate term and price protection. Take multi year only when the price hold and uplift cap are durable.
7. Sequence inside the GenAI calendar. Align the OpenAI renewal with your other AI contracts.

Frequently asked questions