A price per million tokens is a rate, not a forecast of your AI bill. Your actual cost depends on how many input, cached-input, and output tokens a task uses, plus any separately billed tools, modalities, or agent activity—and on the model, service, and agreement behind the rate. The reliable way to budget is to measure representative tasks and calculate cost per successful task using the terms that apply to your account.
Why a token price doesn’t tell you the bill
Token pricing is usually only one part of the calculation. For eligible token-based Enterprise agreements, OpenAI’s rate-card formula adds input, cached-input, and output charges; other applicable feature charges and fees may also apply. The agreement determines relevant discounts and commercial terms. That is a provider-specific example, not a universal billing formula: check the pricing rules for the service you use and the terms attached to your account. OpenAI’s Enterprise rate card describes its scope.
As an Amazon Associate I earn from qualifying purchases.
Even when two models show a lower rate for one category, they may not cost less to complete the same job. Models can tokenize identical text differently and may produce different amounts of output or reasoning tokens. Comparing only input rates—or just the visible length of the final answer—misses those differences. OpenAI’s token guidance cautions that a lower price per million tokens does not necessarily mean lower total task cost. OpenAI Help Center: Understanding and counting tokens.
Which parts of usage can be billed?
Input, cached input, and output
Input is the content sent to the model; output is what it generates. Some services price cached input differently from ordinary input. A prompt’s visible text alone may not reveal the full token count, so use the provider’s reported usage fields when available and apply the rate for each category.
#1 Best Overall
Prompt caching can reduce repeated-input charges
For supported OpenAI API prompts longer than 1,024 tokens, automatic Prompt Caching can reuse the longest matching prefix in 128-token increments. OpenAI says caches are typically cleared after 5–10 minutes of inactivity and removed within one hour after last use. These are OpenAI’s documented behaviors, not a guarantee for other providers or every model. Check the API response’s usage details to confirm how many tokens were actually cached rather than assuming a cache hit. OpenAI Prompt Caching documentation.
For workloads with stable repeated prefixes, test whether reusing that context yields cached usage in practice. A cache’s existence does not establish that a particular request received the lower cached-input rate.
Tools, modalities, and agent loops
Tool use can add cost beyond model token rates. OpenAI’s API pricing lists separate charges for certain tool calls or storage, while tokens used by built-in tools are billed at the chosen model’s rates. Google’s managed-agent pricing, by contrast, describes inference charges that include standard input, output, and intermediate input or reasoning tokens generated during agentic loops; tool fees are handled separately under the relevant pricing rules. The details differ by provider and feature, so inspect the applicable pricing page rather than assuming a single rule. OpenAI API pricing · Google Cloud Vertex AI generative AI pricing.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsPublished rates are examples, not your quote
The following figures are dated illustrations, not a normalized comparison between vendors or a promise of the rate your business will receive. OpenAI published them on its Enterprise rate card, checked October 4, 2026. They apply to eligible Enterprise token-based use; actual discounts and commercial terms depend on the customer’s agreement. Standard-mode rates per 1 million tokens:
| Model | Input | Cached input | Output | Scope |
|---|---|---|---|---|
| GPT-6 Astra | $10 | $1 | $50 | OpenAI Enterprise rate card; eligible token-based agreements, Standard mode |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | OpenAI Enterprise rate card; eligible token-based agreements, Standard mode |
OpenAI’s API pricing page lists a separate short-context table for gpt-6-astra: $10 input, $1 cached input, $12.50 cache writes, and $50 output per 1 million tokens. That page also presents long-context pricing and additional billing details. Do not substitute one rate card for another: confirm the current model, service, context band, endpoint, and account terms that apply to your requests. OpenAI API pricing.
Some conditions are time-sensitive. On the pricing page checked October 4, 2026, OpenAI said eligible regional-processing endpoints for models released on or after March 5, 2026, had a 10% uplift; it also said GPT-5.6 Sol promotional pricing was available at least through November 21, 2026. Eligibility and dates matter, and these details can change. Verify the live pricing page and your agreement before using them in a budget.
How to estimate the cost of your own workload
- Choose representative tasks. Include routine requests and high-volume or otherwise demanding cases from the actual workflow. Track task success as well as spend; a cheap response that fails the job is not a useful cost comparison.
- Capture usage for each task. Record the model, request features, input tokens, cached input, output tokens, and any reported reasoning or tool usage. Use the provider’s own definitions and response fields; OpenAI documents cached-token reporting in API usage details. OpenAI Prompt Caching documentation.
- Apply the rate that matches each category. Use current input, cached-input, cache-write, and output prices where applicable, along with the correct context band. Confirm whether you are using API pay-as-you-go, an eligible Enterprise agreement, a committed tier, a promotion, or another arrangement.
- Add separate charges. Include separately priced tool calls, storage, modalities, and agent-loop activity as the provider’s rules require. Check geography, endpoint, service tier, and account-specific terms.
- Test caching instead of assuming it. For repeated prompts, compare requests with stable prefixes and verify the cached-token count in usage data. A theoretical cache saving is not a measured saving.
- Compare cost per successful task. Weigh spend against quality, latency, context requirements, capacity needs, and contract predictability. There is no verified cross-provider business benchmark here that makes one published rate a universal winner.
A basic calculation for a token-priced request is:
Task token charge = (input tokens × input rate + cached-input tokens × cached-input rate + output tokens × output rate) ÷ 1,000,000
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Use the matching rate units and currency, then add applicable non-token fees separately. This calculation is only as complete as the usage categories and agreement terms you include.
Best Value
What to compare before choosing a model or contract
Compare options on the workload you actually run, not on a single headline number. A useful evaluation records:
- Input, cached-input, cache-write, and output rates, and the token mix a representative task generates.
- How the model tokenizes the same content and how many tokens it needs to complete the task successfully.
- Context-length requirements and any context-dependent pricing.
- Reasoning, agent-loop, tool, storage, and modality charges.
- Task quality, latency, and capacity needs.
- Geographic or service-tier uplifts, discount eligibility, and the term or conditions of a commitment.
Pay-as-you-go versus committed capacity
Procurement terms can change predictability without proving that a commitment is cheaper. OpenAI describes Scale Tier for Enterprise customers as pre-purchased token capacity for a specific model snapshot, with a minimum 30-day term; some models use combined input/output accounting. Review the offer’s capacity, model, accounting method, minimum term, and your expected utilization against your agreement before comparing it with pay-as-you-go. OpenAI Scale Tier.
For scale, OpenAI’s published Scale Tier example prices each GPT-4.1 input unit at $110 per day for 30,000 input tokens per minute, and each output unit at $36 per day for 2,500 output tokens per minute; each unit is purchased for at least 30 days. Those figures describe that specific offer, not a general benchmark or a claim that committed capacity will reduce every customer’s cost.
Why there is no universal “typical business bill”
The provider pricing and help materials cited here establish billing mechanics and published rates, but they do not establish an independent average for business-wide AI spending. A company’s total depends on its workload, model and feature choices, usage mix, account terms, and how much work succeeds on the first attempt. Build an estimate from measured tasks and current account-specific pricing rather than treating any provider’s per-million rate as a market-wide forecast.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
