LLM API bills are based on tokens, not words. A request can incur separate charges for ordinary input, cached input, and generated output; some models also count hidden reasoning tokens as output usage. The reliable way to budget is to measure a representative request, apply the selected model’s current per-million-token rates, and then add tool, search, grounding, storage, or multimodal charges.
There is no universal token price. Tokenization varies by model, encoding, language, and message structure, while pricing changes by model, context length, service tier, region, and cache policy. The examples below show how to calculate a bill without treating a provider’s advertised input rate as your total cost.
As an Amazon Associate I earn from qualifying purchases.
What an LLM API token is—and why token counts are not word counts
A token is a piece of text processed by a language model. It may be a character, a word fragment, a complete short word, or punctuation. The same sentence can produce different token counts with different models or encodings, and languages do not all tokenize at the same rate. API requests can also include system instructions, conversation history, tool definitions, JSON schemas, images, and files, all of which may contribute to usage. OpenAI’s token guidance explains these counting factors in detail: Understanding and counting tokens.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →For budgeting, record the usage fields returned by the exact endpoint you call rather than estimating from character or word counts. A short visible answer can still be expensive if the prompt contains a large context or if the model uses substantial internal reasoning.
#1 Best Overall
- [Ultra-Lightweight Ear Buds Designed for Small Ears] Each earbud weighs only 3.7g and features a compact, ergonomic in-ear design made especially for small ears. Secure, low-profile, and comfortable for workouts, all-day wear or overnight sleeping.The ultra-slim profile sits flush in your ear with no hard edges poking or rubbing, so you can even sleep on your side without discomfort.
- [Immersive Stereo Sound with TOZO OrigX Technology] TOZO OrigX tuning delivers clear vocals, balanced mids, and natural stereo sound for music, podcasts, and videos.
- [Long Battery Life for Daily Use] Get up to 7 hours of playtime on a single charge, with up to 32 hours total using the charging case—ideal for workdays, commuting, and extended listening sessions.
- [Bluetooth 5.3 & Stable Connection] Bluetooth 5.3 provides fast pairing, stable wireless performance, and reduced dropouts as you move around home or office.
- [Deep Bass with Clear Vocals] High-performance drivers produce punchy bass while keeping vocals clean and detailed for everyday listening.
The four token categories that shape an API invoice
| Category | What it represents | How it is usually priced |
|---|---|---|
| Input tokens | Text and other request content supplied to the model, including instructions and conversation context. | Charged at the model’s standard input rate, usually quoted per 1 million tokens. |
| Cached input tokens | A reusable portion of a prompt served from a provider’s prompt cache. | Often has a lower read rate than ordinary input, but eligibility, minimum lengths, expiration, and write or storage charges vary. |
| Output tokens | Tokens generated by the model for its completion or response. | Charged at the model’s output rate. Some providers include hidden thinking or reasoning tokens in this category. |
| Reasoning tokens | Internal tokens used while a reasoning model works through a response; they may not appear in the returned text. | Where the provider reports or includes them, they count toward output usage and use the applicable output pricing. |
Input tokens
Input includes more than the user’s latest message. System prompts, prior turns, retrieved documents, tool instructions, function schemas, and multimodal parts can all increase the request count. If your application resends a long policy or chat history on every turn, that repeated context can dominate the bill even when the user asks a one-line question.
Output tokens
Output is the generated completion, including structured JSON or tool-call arguments when the endpoint counts them as generated content. A response limit can cap output, but it does not make the output free; the tokens actually generated are what matter.
Cached input tokens
Caching reuses previously processed prompt material. A cache hit is still input usage, but it may use a different rate from a cache write or an uncached request. The provider’s definition of a cacheable prefix and its expiration window determine how much of a request qualifies.
Free tools Windows power users keep installed
One-click scans. No signup required.
Reasoning tokens
Reasoning tokens are not necessarily visible in the final answer. OpenAI states that reasoning tokens count toward output usage and are billed as output tokens. Google’s Gemini pricing likewise says its output prices include thinking tokens. Consequently, visible response length alone cannot predict an invoice.
A practical token-cost formula
For a text request with separately reported categories, calculate the token portion as:
Rank #2
- 【IP68 Waterproof Multifunctinal Smart Watch】Torjalph fitness activity tracker smart watch is IP68 waterproof technology ,you can wear it when swimming,handwashing,it's rain-proof which means that you perform multiple exercise modes outside without worrying about sweat or rain corroding your fitness smart watch,and this watch equipped with a 230 mAh large capacity battery magnetic charging ,and this sports activity tracker watch can be used 5-7 days If fully charged.
- 【Larger 1.83'' Full Screen Watch & Multi-dial Selection】1.83 inch full HD color touch screen smart watch for men / women provides excellent 240*284 resolution image quality ,This Bluetooth smart watch offers a better touch experience and remarkable immersion and ,more than 200 watch faces for you to choose from App"Da Fit ", you can change the dial style according to your mood to wear keeping the world on your wrist
- 【Accurate Activity Tracker Smart Watch 】sports watch activity tracker uses sweat and dust resistant breathable silicone strap, our Fitness Tracker Smart Watch for Android iOS Phones supports 12 sports modes,walking,running,bicycle,swimming,badminton,basketball,football,rope skipping. This bluetooth calls smart watch for men women is lightweight and stylish, bringing more convenient and relaxing sports experience while also providing more protection.
- 【Your 24 Hour Personal Daily Health Monitor】Fitness watch for women men full touch screen IP67 waterproof bluetooth smartwatch for Android iOS Phones with heart rate blood pressure monitor multifunction Andriod smart watch with steps counter sleep monitor sports fitness watch, monitoring blood oxygen saturation allows you to understand your physical condition in real time to track fitness progress ,to know your health status in time.
- 【More Functions Bluetooth Call Smart Watch】This fitness activity smart watch full screen for iPhone and Android phones includes a bunch of new features including:The bluetooth smart watch can hang up the phone, receive and display WhatsAPP /Text /Facebook /Twitter /Messenger reminder.Support sedentary reminder, alarm clock, stopwatch, timer,weather information,music control, remote camera control compatible with Android 5.0 and above or iOS 9.0 and above smartphones.
token charge = (input tokens / 1,000,000 × input rate) + (cached input tokens / 1,000,000 × cached-input rate) + (output tokens / 1,000,000 × output rate)
Use the rates for the exact model, tier, region, context class, and billing mode. Then add any charges priced outside those categories, such as built-in tool calls, search or grounding requests, image or audio processing, and cache storage. OpenAI’s pricing documentation notes that endpoints themselves are not separately priced, while some built-in tools and their calls or content can incur additional charges: OpenAI API pricing. Google’s schedule lists separate grounding and storage items, and Anthropic distinguishes cache writes from cache reads in its model table: Gemini API pricing and Claude pricing.
Worked illustration (not a provider quote)
Suppose one workload uses 100,000 uncached input tokens, 900,000 cached input tokens, and 50,000 output tokens. Apply hypothetical rates of $2.00 per million ordinary input tokens, $0.20 per million cached input tokens, and $10.00 per million output tokens:
| Usage | Hypothetical rate | Calculation | Charge |
|---|---|---|---|
| 100,000 input tokens | $2.00 per million | 0.1 × $2.00 | $0.20 |
| 900,000 cached input tokens | $0.20 per million | 0.9 × $0.20 | $0.18 |
| 50,000 output tokens | $10.00 per million | 0.05 × $10.00 | $0.50 |
| Total token charge | $0.88 | ||
These rates and quantities are illustrative arithmetic, not a current price from any provider. Replace them with measured usage and the live rate row for your selected model before making a budget commitment.
How prompt caching changes the calculation
“Cached” does not mean one universal discount. Providers differ in what they cache, how long a cache entry lives, whether the first request is a paid write, and whether storage is charged separately.
Rank #3
- [Low Noise High Fidelity] New bee guitar cord with 22AWG oxygen-free-copper center conductors provides enhanced signal transfer, the PVC insulation material and the high-density stranded copper double shield provide noise rejection for quiet operation
- [Ideal for Choice] New bee 1/4 straight and right angle metal connectors, ensures a secure connection in any instrument, pedal, or amp jack. L shape great fro on stage guitar stands
- [Wide Compatible] Suitable for mixer, power amplifier, microphone, effector, decoder, equalizer, electoronic organ, electric guitar, electric bass and other electronic equipments. Widely used in the installation of smart home, karaoke speaker, professional stage audio, etc
- [Easy Disassemble] Making incidental repairs quick and easy! Very nice, flexible cable
- [Dont Hesite to Order] 24 hours customer services and professional technology team are standing by
OpenAI automatic prompt caching
OpenAI documents automatic prompt caching for supported requests longer than 1,024 tokens. The service can reuse the longest previously computed prefix, beginning at 1,024 tokens and extending in 128-token increments. Response usage includes a cached_tokens value, which lets you measure how much input received cached treatment. Details and supported models are described in OpenAI prompt caching. The pricing table still determines the applicable uncached and cached-input rates.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Anthropic cache writes and hits
Anthropic’s pricing table separates 5-minute cache writes, 1-hour cache writes, and cache hits or refreshes. A workload that writes a prompt once and reuses it frequently can have a different cost from one that continually creates short-lived entries. The page also states USD rates and conditions related to prompt length and inference geography, so those conditions must be checked with the chosen model: Anthropic Claude pricing.
Google Gemini caching and additional services
Google’s Gemini schedule presents model and tier-specific token prices and lists separate storage and grounding charges. Do not assume that a cached context, a grounding request, or a multimodal part is covered by the basic text input line. Check the current Gemini schedule for the selected model and region: Google Gemini API pricing.
When caching saves money
Estimate the reuse rate, not just the theoretical discount. A cache is useful when the same prefix is sent repeatedly and remains eligible for the provider’s retention window. If users frequently alter text near the beginning of a prompt, the reusable prefix may be short. Track uncached tokens, cached tokens, cache writes, cache reads, and expiration-related refreshes separately in your usage data.
Reasoning tokens: the hidden part of output usage
Reasoning models may spend internal tokens planning, checking constraints, or solving a problem before returning visible text. Those tokens can be billed even though they are absent from the answer body. OpenAI’s Help Center gives the key warning: A lower price per million tokens does not necessarily produce a lower total cost: models can tokenize the same text differently and generate different amounts of output or reasoning.
The statement appears in OpenAI Help Center’s token guidance.
Rank #4
- Digital Family Calendar: BSIMB digital calendar wall touch screen puts an end to your chaotic life. Designed for busy families, the digital calendar keeps you organised and in step with your family. Assign different colours and tasks to each family member to make family scheduling easier and more intuitive.
- Easy Setup and Auto-Sync: Plug in, connect to WLAN and auto-syn your electric calendar in minutes. Digital wall calendar compatible with Google calendar, iCloud calendar, Outlook calendar, Cozi calendar and Yahoo calendar. Add events directly on the device or through the ecalendar app. Electronic calendar stays current whether you're at work or traveling—no manual updates needed.
- All-In-One Home Assistant: Smart calendar builds habits and encourage independence with personalised chores. Display daily meal plan to reduce the daily ‘What's for dinner?’ stress. To-do lists keep track of everything you want. Turning chores into valuable achievements with stars and rewards, making life easier and more enjoyable.
- Manage Anytime Anywhere: BSIMB electronic calendar for family stay connected with the powerful ecalendar app. Access and manage your digital wall calendar anytime, anywhere with free features like stars & rewards, photo screensaver, chore charts, meals plans, to-do lists etc. Electronic planner paid features like Magic Import tailor advanced tools for users with special needs.
- Stay in Touch with Loved Ones: Share photos or videos from your phone to electronic wall calendar with app, instantly update your life moment, and keep in touch with your family, especially those who are far away; you can also surprise your smart calendar digital wall by sending him a wish card; it's also great for keeping tech-illiterate elders in touch with the youngsters.
For a fair estimate, capture the provider’s reported output and reasoning fields when available. If a provider includes thinking tokens in its output price without exposing a separate count, use the total output usage supplied in the response or billing export. Do not infer reasoning cost from the number of words a user sees.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What current provider schedules actually show
Price schedules are snapshots, not permanent market statistics. Model names, tiers, regions, and rates can change, so quote a figure with its provider, model, unit, tier, and date checked.
| Provider and source | Documented pricing structure | Published figure or qualification |
|---|---|---|
| Google Gemini | Model- and tier-specific input and output rates, with distinct modality, grounding, and storage items. | Gemini 3.1 Flash-Lite Standard listed text/image/video input at $0.25 per 1 million tokens and output at $1.50 per 1 million tokens, checked 2026-10-13. This is a schedule snapshot for that model and tier. |
| OpenAI | Separate model-specific columns for input, cached input, and output; tool charges may be listed separately. | Rates are not stated here because the applicable row depends on the current model, region, and product. Use the live OpenAI pricing table. |
| Anthropic Claude | USD model rates with separate input, output, 5-minute cache-write, 1-hour cache-write, and cache-hit or refresh columns. | Rates are not stated here because the selected model, prompt-length conditions, and inference geography determine the applicable row. Use the live Anthropic pricing table. |
The Gemini numbers above were checked on 2026-10-13 and can change. The OpenAI and Anthropic pages should be consulted immediately before publication or procurement because API pricing is volatile.
How to estimate the cost of your own workload
- Define one representative task. Include the normal system prompt, typical user message, retrieved context, tool schemas, files, and expected response format.
- Run enough samples to capture variation. Measure short and long prompts, successful and retried calls, and easy and difficult tasks. Record input, cached input, output, and reasoning usage fields returned by the API.
- Separate cache states. Mark first-use cache writes, cache hits, refreshes, and requests that did not qualify. Calculate the observed reuse rate.
- Apply the exact rate row. Match model, context-length band, service tier, batch or flex mode, region, and modality. Convert every usage count to millions before multiplying.
- Add non-token charges. Include search or grounding requests, built-in tools, cache storage, image or audio units, and any other item listed outside token pricing.
- Project volume. Multiply the per-task result by expected calls, then add a separate allowance for retries, safety rechecks, and traffic spikes. Keep this allowance visible rather than hiding it in a lower average.
- Recheck before launch. Confirm that the model and price page still show the same rates and eligibility rules on the day you approve the budget.
How to compare providers fairly
Use the same workload and accounting boundaries for every candidate. A useful comparison records the following for each provider and model:
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →- Measured ordinary input tokens for the identical prompt and context.
- Measured output tokens, including reported or included reasoning usage.
- Cache eligibility, write cost, read or hit cost, retention period, and expected reuse percentage.
- Context limits and any price multiplier for long prompts, service tiers, batch or priority processing, or inference geography.
- Separate charges for tools, search, grounding, storage, images, audio, and other modalities.
- Task quality, completion success, latency requirements, retries, and the number of calls needed to finish the job.
A lower per-million-token rate can still produce a higher task bill if that model tokenizes your content differently, generates longer answers, reasons more, needs retries, or requires extra calls to reach the same quality. OpenAI explicitly recommends testing representative tasks instead of comparing only visible response length: token guidance.
Common budgeting mistakes and how to avoid them
Using word count as a token forecast
Word-to-token ratios vary by language, punctuation, code, and model. Count tokens with the tokenizer or usage fields for the actual model and payload.
Pricing only the visible answer
Include the complete request context and any hidden reasoning usage. A concise answer may follow a large prompt or extensive internal computation.
Applying a generic cache discount
Cache policies differ in thresholds, prefix matching, expiration, write pricing, read pricing, and storage. Use provider-specific usage fields and rates.
Recommended Free Tools
Ignoring non-token line items
Search, grounding, tool calls, storage, images, audio, and files can sit outside the basic input/output formula. Add each item from the selected provider’s schedule.
Comparing different tiers or regions
Standard, batch, flex, priority, long-context, and regional options may have different rates or conditions. Compare like with like and document the assumptions.
Treating a dated rate as permanent
Write the check date next to every quoted figure and re-open the provider’s current pricing page before signing off on a forecast.
A repeatable budget worksheet
For every model under consideration, keep one row per workload and record:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Model name, endpoint, region, service tier, and pricing-page check date.
- Average and high-percentile ordinary input tokens.
- Average cached input, cache writes, cache hits, and observed reuse rate.
- Average output and reasoning usage, or the provider’s combined output field.
- Tool, grounding, storage, image, audio, and other non-token units.
- Cost per successful task, retry rate, monthly task volume, and projected monthly total.
This worksheet turns “How much do LLM API tokens cost?” into a workload-specific answer. It also makes a future price or model change easy to recalculate without confusing a new schedule with a change in application behavior.
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.
