Claude vs OpenAI Enterprise: What the Seat Price Leaves Out

A seat price is only one part of an AI contract. The usage it includes, the tools it unlocks and the terms attached to overage can change the final bill.

Anthropic charges $20 a month per Enterprise seat, billed annually, with usage billed separately at API rates. Its minimum is 20 seats for self-serve and 50 for sales-assisted plans. OpenAI offers standard ChatGPT seats and, on eligible credit-based agreements, Codex-only seats without a fixed per-user fee. Codex-only access does not include ChatGPT.

Start with equal access. Then test the contract your team can actually buy.

Usage can outweigh the seat

Across enterprise Claude Code deployments, Anthropic reports an average of $150 to $250 per developer per month, with 90% of users below $30 per active day. Those are deployment-wide figures, not a Sonnet-specific forecast.

For an illustrative 1,000-person software company, assume 400 developers at the $200 midpoint and budget $15.70 for each other employee. The latter is Ramp's median firm AI spend per employee, not a measurement of chat consumption. On Claude, the resulting annual scenario is $1.31M, including $20k in monthly seats and $89k in assumed monthly usage.

Seats are 18% of this scenario, not a universal share of enterprise spending.

The table varies the usage budget. Its $660 heavy-user scenario multiplies $30 by 22 active days; it is not a measured 90th-percentile monthly bill. The $1,500 scenario uses Uber's reported monthly cap, not its employees' typical spend.

Equal token prices do not guarantee equal bills

On the published rate cards, Claude Sonnet 5.5 and GPT-6.1 Sol both cost $2 per million input tokens and $10 per million output. Under the same assumed usage, metered in full, and a standard chat seat for everyone, the 1,000-person scenario comes to $1.69M for OpenAI and $1.31M for Claude. OpenAI's seat figure is reported, not a quote, and included usage is not deducted.

Model choice changes the scenario. Move the developers from Sonnet to Opus 5.5, at $4 and $20 per million input and output tokens, and the estimated Claude bill rises by $0.96M a year. Artificial Analysis reports weighted benchmark costs of $10.02 per task for Sonnet 5.5 and $1.27 for GPT-6.1 Sol in the measured configurations: 7.9 to one.

Applying that benchmark ratio mechanically produces an illustrative $0.85M OpenAI scenario. It is not a prediction: benchmark tasks, model configurations and completion quality may differ from your work.

Measure your own tasks before treating that ratio as a saving.

Included usage can change the comparison

OpenAI's credit-based standard Enterprise seats include baseline access; additional usage can draw on credits. Token-based agreements have different terms. Claude's current Enterprise seat does not include usage. Whether a bundled allowance covers a particular workload must be checked against the actual agreement and usage limits.

For a 2,000-person media company using the same illustrative per-person budgets, the fully metered scenarios are $1.30M for Claude and $2.06M for OpenAI. The planner can instead assume non-developer OpenAI usage is included, but cannot establish whether the workload fits.

For a 50-person agency, the seat-only Claude Team scenario is $40,800 a year, with Premium seats for developers; its metered Enterprise scenario is $87,768. Team overages are excluded, so these are not equivalent complete-usage forecasts. ChatGPT Business also includes usage up to plan limits, with additional eligible usage paid in credits under its terms.

Twelve months buys seats, not a budget

Claude Enterprise bills seats at the start of an annual term and prorates additions; self-serve seats cannot be removed mid-term. In this scenario the 12-month seat fees are $240,000 for Claude and $618,000 for OpenAI. The OpenAI figure uses a reported price and term, and neither figure captures any separately negotiated usage commitment.

Usage can stop at an allowance or credit limit, or continue as an overage, depending on the plan and admin settings. Both vendors offer controls; confirm the defaults and caps in the contract.

Pilot with a representative group, measure cost per completed task, and ask each vendor for the same access and workload. A quote with different included usage or a negotiated commitment could change the result.