The three pricing models
- Seats
Per-user subscriptions
Access to assistants such as ChatGPT Enterprise, Claude Enterprise, Microsoft 365 Copilot and Gemini, billed per person, usually on annual terms.
- Usage
Tokens and credits
APIs and agents are billed on what they process, in tokens or in vendor credits such as Microsoft's Copilot Credits. Rates differ by model, and new models arrive with new rate cards.
- Commitment
Committed spend
Prepaid credits, savings plans and reserved capacity lower rates in exchange for a spend commitment. Committed spend is paid whether or not it is used.
How the major providers sell it
- OpenAI
ChatGPT Enterprise and the API
ChatGPT Enterprise is priced per customer, with no published list price. The API is billed per token, with input, output and cached tokens priced separately.
- Anthropic
Claude Enterprise and the API
Claude Enterprise lists at $20 per seat per month, billed annually, with usage billed separately at API rates and no included token allowance. Admins can set spend limits by organization and user.
- Microsoft
Microsoft 365 Copilot and Azure OpenAI
Copilot is $30 per user per month, and since September 2026 advanced AI work draws usage-based Copilot Credits. Azure OpenAI is billed per token or through reserved capacity.
- Google
Gemini Enterprise and Vertex AI
Gemini Enterprise is sold per seat or pay-as-you-go, and since August 2026 through Flexible Savings Plans at 10% off for one year or 20% for three.
- AWS
Bedrock
Bedrock is billed on usage, with batch processing and provisioned throughput as alternatives. How Bedrock spend counts toward an AWS agreement depends on the agreement.
What makes AI contracts different
A traditional software contract is priced on something stable, like seats or cores. AI is priced on activity. An assistant used lightly costs little beyond the seat. An agent that runs a multi-step workflow can consume far more than anyone forecast, and coding and agent tools have become the fastest way for AI budgets to run over.
Terms also move faster than in most software. Models are released and retired within a contract term, rate cards change with them, and seats bought ahead of adoption often go unused. Data use, model changes and price protection all matter at signature, and AI spend increasingly sits inside larger cloud and software agreements rather than on its own.
Where Deal IQ comes in
This page covers how the pricing works and where the cost builds. Getting a better price on your own contracts is the negotiation itself. Deal IQ negotiates AI contracts across OpenAI, Anthropic, Microsoft, Google, AWS and the AI features now bundled into enterprise software, as part of $20B+ in contract value negotiated across 3,000+ negotiations with 600+ vendors. Our negotiators work with a network of 350+ former vendor sales, pricing and deal desk leaders, and we negotiate directly with the vendor or behind the scenes with your team. Our fee comes from the savings.
Frequently asked questions
How is enterprise AI priced?
Through a mix of per-seat subscriptions, usage-based pricing in tokens or credits, and committed spend. Most enterprise AI agreements now combine all three.
How much does Claude Enterprise cost?
Claude Enterprise lists at $20 per seat per month, billed annually, with usage billed separately at API rates and no included token allowance.
Does ChatGPT Enterprise have a published price?
No. OpenAI prices ChatGPT Enterprise for each customer.
Why do AI costs run over budget?
Because usage, not seats, drives the bill. Agents and coding tools can consume far more than forecast, and seats bought ahead of adoption often go unused.
THE DEAL IQ TREND REPORT
Track what vendors are changing.
Pricing, packaging, sales priorities and commercial terms, read from the other side of the table.
Get the latest issue and subscribe to future reports.
DEAL IQ IN NUMBERS