← All posts

Guides

How Do You Verify Your Negotiated AI API Price?

Verify that your enterprise AI API quote is active for the right account, model and usage before scaling production traffic.

To verify a negotiated AI API price, match the written quote to the production account and exact model, confirm the currently applicable rate, then check a settled test charge. A signed discount or a cheaper public price does not establish that your application is using the agreed commercial terms.

An enterprise can finish price negotiations and still have an unresolved implementation question: has the agreed rate reached the account our application actually uses? This article addresses that handoff from procurement to engineering and finance. It is a documentation-based acceptance method, not a report of a customer billing incident or a completed pricing test.

At Token360, enterprise procurement covers text, image, audio and video models. The enterprise offering includes custom pricing; the actual scope must be agreed for the engagement. This guide checks whether those agreed terms are observable in the production setup.

What counts as proof? · Account scope · Rate interpretation · Test charges · Sign-off record

What evidence shows that an enterprise API price is active?

A useful acceptance record connects three things: the approved commercial terms, the price returned for the intended production account, and a settled charge from a representative request. Each answers a different question. The agreement states what should happen; the account record and charge show what the implementation currently does.

EvidenceWhat it establishesWhat it does not establish
Approved quote or price scheduleThe model scope, currency, units and agreed conditionsThat every production credential has the correct account association
Authenticated pricing snapshotThe price presented for that account and model at the check timeA historical price or a guarantee that every possible request has the same rate
Settled test chargeThe recorded cost of one identified executionCorrect pricing for untested models, settings or future periods

Use the three records together. A successful API response alone verifies neither a discount nor a commercial agreement. Conversely, a different-looking rate field need not mean an error: one system may expose a base rate and factor, while another displays the payable rate directly.

Start with the rows of the agreement you intend to use at launch. If the quote covers several models, record an acceptance status for each relevant row instead of giving the entire account a blanket “discount verified” label.

Which account should you use to check the negotiated price?

Check the account that will pay for production inference, using a credential associated with that account. Do not substitute a personal test account, a colleague’s login or an unrelated workspace. Ask the application owner to confirm the production configuration without exposing the secret value of the API key.

Token360’s model-pricing reference, checked November 19, 2026, says the returned price belongs to the account owning the authenticated key. An enterprise key can return its negotiated price; a personal key does not inherit another enterprise’s terms.

The buyer’s handoff should name the enterprise account, the application or environment, the credential’s non-secret display identifier, and the exact model ID. Engineering should confirm that these match the deployed configuration. A procurement spreadsheet cannot establish which credential a running service uses.

If a pilot used a personal account, treat the move into the enterprise account as a new acceptance check. For group structures, keep this separate from whether subsidiaries are covered by one API contract: contractual coverage and the account used by an application are related decisions, but one does not prove the other.

How do you distinguish a discount factor from the final rate?

First identify whether the displayed number is a base rate or an already negotiated payable rate. Apply only the adjustment the pricing record calls for, once. Then preserve the unit and currency. Applying a percentage to an already discounted figure creates an artificially low expectation before any inference request is sent.

For Token360, GET /v1/models/{model_id}/pricing reports current account pricing. When price_factor is numeric, multiply the rate by it; when null, use the rate. discount_percent_off restates the factor rather than adding another discount. The reference also requires selecting the first matching pricing rule. Its as_of timestamp describes the current response, not a price history. Source: model-pricing reference, checked November 19, 2026.

What would a double-discount mistake look like?

Hypothetical arithmetic only—not Token360 rates or measured savings. Suppose an approved example has a base input rate of $2.00 per million tokens and a factor of 0.80. The payable input rate is $1.60 per million. Reducing $1.60 by another 20% would incorrectly predict $1.28.

Illustrative recordCorrect interpretationIncorrect interpretation
$2.00 base rate × 0.80 factor$1.60 per million input tokens$1.28 after subtracting the same discount twice
$1.60 negotiated rate; no further factor$1.60 per million input tokensAssuming another reduction merely because the quote mentions a discount

Ask finance to compare payable rates, not unlike fields. If the agreed quote changes over time, use the separate guide to inference price changes and effective dates. This acceptance check does not replace the agreement’s rules for future changes.

What must match besides the model name?

The comparison must preserve the conditions that determine which rate applies: model ID, billable unit, relevant generation settings, account scope and effective period. Only include additional conditions when they exist in the agreement or pricing record. Do not assume that a model-family name identifies every billable configuration.

Prepare one short row for each commercially distinct configuration you plan to use. For a language workload, separate input and output rates rather than inventing an average token rate. For media, carry the quoted resolution, duration or output unit into the acceptance record where relevant.

Token360’s billing documentation states that meters vary by model and that a public upstream price is not automatically the Token360 route price. Therefore, a vendor’s public pricing screenshot is context, not proof of the amount your account should pay. Documentation checked November 19, 2026.

Keep this exercise focused on applying the agreed terms. The separate AI API volume-discount guide explains what can be discounted; reasoning-token billing covers hidden generation usage. Neither should be replaced by a single “percentage off” field in a launch checklist.

How should you check a small test charge?

Use a small, representative request from the intended production account, keep its returned identifier and settings, and compare the settled charge with the agreed payable rates and recorded usage. The purpose is to check one pricing path before rollout. It is not a quality benchmark or proof of every possible billing combination.

  1. Capture the expected configuration. Record the model, account, applicable price schedule, units and check time before the request.
  2. Run a bounded request. Use non-sensitive input and an appropriate output limit. A test can incur normal usage charges; a zero-cost request may not exercise the negotiated rate.
  3. Use actual metered usage. An output-length limit is not the number of output tokens generated. Include other applicable components rather than forcing a two-rate formula onto every model.
  4. Wait for settlement. A missing or pending charge is unresolved evidence, not proof that usage was free.
  5. Compare and record the result. Keep matched, mismatched and pending cases separate. Ask the billing team to explain discrepancies before treating the row as accepted.

In a deliberately simplified example, 1,000 input tokens at $1.60 per million and 200 output tokens at $4.00 per million produce $0.0024 in model usage charges. These invented rates illustrate multiplication only; they exclude additional meters and any provider-specific precision or rounding rules.

For Token360, the request-usage endpoint returns settlement and charge information for an execution. Prefer the platform-returned X-Request-ID for lookup; recorded usage varies by model. Documentation checked November 19, 2026. A broader month-end workflow belongs in request-level API billing reconciliation, rather than in this price-activation checklist.

What should you do if the price does not match the quote?

Keep the affected pricing row unapproved while checking account scope, model configuration, rate interpretation and settlement status. Send a concise discrepancy record to the commercial and technical owners. A mismatch needs an explanation; it does not, by itself, establish model substitution, intentional overcharging or a platform-wide billing fault.

Provide the quote reference, enterprise account identifier, model ID, UTC timestamps, expected rate calculation, returned request ID and observed charge. Do not send API-key secrets or unrelated customer prompts. State whether the discrepancy is a different unit rate, different usage amount or an unsettled record; these require different investigations.

Agree on the next action: correct a configuration, clarify the quotation, wait for settlement, or investigate billing. Record any agreed adjustment separately rather than assuming a later credit will appear. Retest the affected configuration after the explanation or change. Unaffected, already accepted rows can retain their evidence.

A price discrepancy is also different from the question of whether a less expensive route delivers acceptable outputs. Use the cheaper-API quality evaluation for that decision, with matched workloads and acceptance criteria.

What should procurement sign off before production scales?

Sign off a defined set of models and commercial conditions, supported by dated account-price snapshots and settled test records. Name an owner for unresolved exceptions. The result should say what was checked, when it was checked and what remains untested, so another team can understand the acceptance without repeating the negotiation.

Acceptance fieldWhat to record
AgreementQuote version or schedule reference and applicable effective period
ScopeEnterprise account, application/environment, non-secret credential identifier, model ID and relevant settings
Expected priceCurrency, billing unit, payable rate and any applicable condition
Observed evidencePricing-check timestamp, returned request ID, recorded usage and settled amount
DecisionAccepted / unresolved / not tested; explanation and responsible owner
Recheck triggerChanged account association, model configuration or commercial schedule

This is a proposed internal acceptance record, not a claim that Token360 automatically produces this document. Procurement owns the agreed terms, engineering confirms the configuration, and finance checks the arithmetic. One person can hold multiple roles, but each question still needs an answer.

For a Token360 engagement, send the model list and intended enterprise account to your commercial contact, and ask how to confirm that the agreed rates are active. Leave dedicated capacity, support and compliance commitments to the applicable agreement. A verified unit price establishes a pricing path; it does not establish those separate service commitments.

  • Enterprise AI
  • API

Bring your next idea to life.

Create a video scene in Creator, or explore models for images, audio, and text to find the right capabilities for your project.

Get started