Claude API pricing is usage-based: Anthropic charges separately for input and output tokens, with rates that vary by model. Output tokens usually cost several times more than input tokens, so response length, retries and tool calls can dominate a bill. A reliable estimate multiplies your measured token counts by the model’s current per-million-token rates, then adds prompt-cache writes and reads, tools, routing premiums and expected retries.
The prices below reflect Anthropic documentation accessed October 16, 2026. Model prices, availability and multipliers can change; verify the live pricing page and your account settings immediately before committing a budget.
As an Amazon Associate I earn from qualifying purchases.
Claude API price snapshot
Anthropic lists prices in US dollars per million tokens (MTok). Input and output are metered independently. The input rate applies to tokens sent in a request; the output rate applies to tokens generated in the response.
| Model | Input price | Output price | Documented qualification |
|---|---|---|---|
| Claude Fable 5.1 | $10/MTok | $50/MTok | List price in Anthropic’s current table |
| Claude Opus 5.5 | $4/MTok | $20/MTok | List price in Anthropic’s current table |
| Claude Sonnet 5.5 | $2/MTok | $10/MTok | List price in Anthropic’s current table |
| Claude Haiku 5.5 | $0.10/MTok for prompts up to 100,000 tokens; $0.50/MTok above 100,000 | $0.50/MTok for prompts up to 100,000; $2.50/MTok above 100,000 | The threshold changes both input and output rates |
| Claude Sonnet 4.6 | $3/MTok | $15/MTok | List price in Anthropic’s current table |
| Claude Haiku 4.5 | $1/MTok | $5/MTok | List price in Anthropic’s current table |
These are list prices, not a promise of a fixed future rate. Anthropic also notes that Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text, with the actual increase depending on content and workload shape. A spreadsheet based only on character counts can therefore understate cost.
#1 Best Overall
- Compact & Portable Design: This Small Excel, Word, PowerPoint, & Windows Cheat Sheet Mouse Pad measures 11.8" x 9.8", offering a portable mouse pad. Its compact size fits in bags or laptop cases, suitable for professionals, students, or travelers.
- Multi-Software Shortcut Guide: Featuring essential shortcut keys for Excel, Word, PowerPoint, & Windows, this small mouse pad provides a quick reference for frequently used shortcuts, offering a reference designed to support workflow efficiency
- Smooth & Precise Surface: The finely-textured surface ensures precise mouse control, allowing for smooth and accurate movement during work or gaming sessions.
- Durable & Non-Slip Design: This Multi-Software Shortcut Mouse Pad features stitched edges to reduce fraying. The non-slip rubber base helps keep it securely in place, and the water-resistant fabric allows for easier maintenance.
- Clear Print Quality: Displays high-resolution printing intended to remain legible through regular use and cleaning
How to compare the headline rates
For a quick sensitivity check, price the same hypothetical workload against each model. One million input tokens plus 100,000 output tokens would cost $15 on Fable 5.1, $6 on Opus 5.5, $3 on Sonnet 5.5, $4.50 on Sonnet 4.6 and $1.50 on Haiku 4.5, before any caching or tools. Haiku 5.5 would cost $0.15 if its prompt remains within the 100,000-token tier; its rates change to $0.50 input and $2.50 output per MTok when that prompt threshold is exceeded. These calculations show why a model comparison must use your actual input-to-output mix rather than comparing input rates alone.
The cost equation you can reproduce
Calculate one request first, then scale it to your traffic:
estimated_cost = (input_tokens / 1,000,000 × input_rate) + (output_tokens / 1,000,000 × output_rate) + cache_write_cost + cache_read_cost + tool_or_server_charges
For a monthly forecast, multiply the request estimate by expected requests and add separate allowances for retries, validation calls and peak-period overhead. Keep input and output counts in separate columns; combining them hides the fact that output is often the expensive part.
Measure tokens instead of guessing
Anthropic’s Token Counting API lets you count a message before sending it. The Messages, Message Batches, Token Counting and Models APIs are available through the direct REST service at https://api.anthropic.com. Use the /v1/messages/count_tokens endpoint in a preflight or budgeting job, especially when prompts include long documents, tool definitions or conversation history.
Token counting is especially useful at product boundaries. Count an uploaded file before accepting it, count the full conversation before appending another turn, and count a batch payload before submission. Keep the measured count with the request identifier so a later invoice reconciliation can distinguish a genuine rate change from a prompt that grew unexpectedly.
Worked Sonnet 5.5 example
Suppose one request contains 100,000 input tokens and produces 20,000 output tokens. At Sonnet 5.5’s listed rates, the arithmetic is (0.1 × $2) + (0.02 × $10) = $0.40 before caching, tools or other modifiers. If that pattern runs 10,000 times in a month, the unmodified token estimate is $4,000; add your measured retry and tool usage rather than assuming every request succeeds once.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
A smaller interactive example illustrates why output limits matter. A request with 10,000 input tokens and 2,000 output tokens on Sonnet 5.5 costs (0.01 × $2) + (0.002 × $10) = $0.04. If an unconstrained response doubles to 4,000 output tokens, the request becomes $0.06 even though the prompt did not change. At high volume, a response-size regression can therefore cost more than a modest increase in prompt length.
Prompt caching: when it lowers the bill
Prompt caching prices the cache operation separately from ordinary input. Anthropic documents a five-minute cache write at 1.25 times the model’s input rate, a one-hour write at 2 times the input rate, and cache reads generally at 0.1 times the input rate. Anthropic states that the five-minute option pays back after one read and the one-hour option after two reads, assuming the cached prefix would otherwise be sent at the normal input rate.
Rank #2
- [MULTIFUNCTIONAL]You'll get 2 pieces computer monitor memo boards that you can stick on the left and right edges of your monitor, and they're the perfect office desk organizers and accessories. Computer monitor side panels desktop organizer are suitable for home work or office,bringing convenience. Desktop memo is used to organize meeting memos, important messages, business cards, planning notes.Paste on the message board to keep track of important things and to-do items to prevent forgetting.
- [🌟HIGHLY QUALITY] The material of computer screen side note holder is transparent acrylic. Durable, simple, stylish, light weight, easy to use, not easy to fall off or break. This cute office supplies for women desk can be used for a long time. This computer desk accessories is waterproof and dirt resistance, and look simple and stylish. The transparent acrylic sticky note holder as cubicle accessories is easy to notice the context of your sticky notes.
- [📋Easy to use] Office must haves cool office gadgets for desk ready to tear, easy to install and remove, not easy to leave traces. You only need to peel off the protective film on the surface of the computer side board memo, wipe off the dust on the edge of the computer monitor, and then stick the desk essentials for women office on the right or left side of the tape, and you're done. A perfect gift for your colleagues, friends or classmates and family members or relatives
- [🏢MULTI-SCENE USE] This desk supplies computer memo board can be applied to home and office, clear your office decor for women, suitable for most computer monitors, screens and cabinets, you can put it where you think, this cute office decor serve as a reminder. Stick on the computer side. It’s a good office gadgets can remind work improve office productivity. Pasted cabinets, dressers, refrigerators, walls, etc as cubicle accessories. To make life more orderly.
- [💌NOTE] The adhesive force of the computer sticky note holder is very strong. It can not be directly pasted on the computer screen. It should pasted on the black edge of the screen. Narrow edge not recommended!!! If you are not satisfied with your purchase, or if the product is damaged or broken in transit, please let us know immediately. We will promptly solve your problem.
| Model | Five-minute write | One-hour write | Cache read |
|---|---|---|---|
| Claude Fable 5.1 | $12.50/MTok | $20/MTok | $0.25/MTok |
| Claude Opus 5.5 | $5/MTok | $8/MTok | $0.20/MTok |
| Claude Sonnet 5.5 | $2.50/MTok | $4/MTok | $0.10/MTok |
| Claude Sonnet 4.6 | $3.75/MTok | $6/MTok | $0.30/MTok |
| Claude Haiku 4.5 | $1.25/MTok | $2/MTok | $0.10/MTok |
Haiku 5.5’s supplied price entry specifies its two prompt-length input and output tiers but does not state cache prices, so do not infer a cache rate for it from another model.
Cache calculation example
Take the 100,000-token Sonnet 5.5 prefix in the earlier example. A five-minute write costs 0.1 × $2 × 1.25 = $0.25. Ten reads cost 10 × 0.1 × $0.10 = $0.10. The cached prefix therefore costs $0.35 in write and reads, instead of charging the normal $0.20 input amount 11 times ($2.20). Calculate the changing, uncached portion of each request normally; only the stable prefix receives cache treatment.
Recommended Free Tools
The break-even rule depends on reuse, not merely prefix size. If a prefix is written once and read once during five minutes, its total cache charge is 1.35 times the normal input price, while the uncached alternative would charge it twice. The first read is enough to recover the write premium. With a one-hour write, the 2-times write charge needs two reads to beat repeatedly paying the normal input rate. A cache-hit report that shows many writes but few reads is a signal to shorten the cached prefix, extend the reuse window, or remove caching from that route.
What belongs in a cache
- System instructions that remain unchanged across requests.
- Tool definitions and schemas reused by many calls.
- Large reference documents or a stable knowledge prefix.
- Conversation history that is repeatedly sent within the cache lifetime.
Cache only data that is actually reused. A prefix that is written once and never read again adds a write charge without creating a saving. Keep changing user text after the stable prefix so a new request does not invalidate the reusable portion. Track cache creation and cache read tokens separately; a high cache-write rate can indicate that your supposedly stable content is changing too often.
Batch API discount for asynchronous work
Anthropic’s Batch API gives asynchronous workloads a 50% discount on both input and output tokens. It is suited to work that can wait for completion, such as offline classification, enrichment, backfills and evaluation runs. It is not a drop-in latency optimization for an interactive chat path: design your queue, completion polling and failure handling around asynchronous results.
Apply the discount to the token components only. Cache charges, server-side tools, routing choices and other documented feature costs should remain separate lines in your forecast unless Anthropic explicitly states otherwise. A useful comparison is to calculate an online scenario and a batch scenario with identical token counts, then add the operational value of waiting for completion. If the job can be deferred, the 50% token reduction can be more material than switching between two adjacent model tiers.
Extra multipliers and charges
The token rates in the table are the starting point. Anthropic documents additional cost or multipliers for several features.
Long-context and prompt-size tiers
Some models apply different rates when a prompt crosses a token threshold. Haiku 5.5 is explicit: prompts up to 100,000 tokens use $0.10 input and $0.50 output per MTok, while prompts above 100,000 tokens use $0.50 input and $2.50 output. Keep a prompt-length distribution, not just an average, so a small number of very long calls does not disappear inside a blended rate.
For example, a workload with a 95,000-token median can still have a costly tail if a few requests include full conversation archives. Record percentile counts and the number of requests above each threshold. If practical, summarize or split those outliers before routing them to a higher tier; otherwise reserve a separate budget for them.
Rank #3
- Stay Organized with Ease: Our To Do List Notepad provides multiple sections with plenty of space to jot down all of your important tasks.A versatile office supply perfect for daily task management, helping you keep track of everything that needs to be done and prioritize effectively.
- Achieve Your Productivity Goals: Begin each day on a positive note, reminding you of your potential, strength, and the significance of staying focused on your goals. Our Daily Checklist Notepad is the ultimate tool to streamline your daily planning and elevate your productivity. With a clear and concise overview of your tasks, you can effortlessly prioritize what truly matters and make consistent strides towards achieving your goals.
- Reliable and Stylish: Our to do list planner combines functionality with durability. With a transparent PP cover, the inner pages are well-protected from dirt and wear. Additionally, the back cover is crafted with thick paperboard, offering a stable surface for your writing needs. Each notepad includes 52 pages of high-quality, 100gsm paper, which is thick and non-bleeding, resulting in a smooth and enjoyable writing experience.
- Versatile Use: This daily checklist notepad, suitable for teacher supplies and a wide range of applications, proves invaluable in various settings including work, school, and home. Whether you need to organize your daily tasks, plan a project, or create a grocery list, this notepad serves as the ideal tool. Its compact size ensures easy portability, allowing you to conveniently carry it with you throughout your day.
- Perfect Present for Anyone: Our Daily To-Do List Notepad is the perfect present for those who value organization and productivity. With its stylish design and versatility, this office desk accessory is also a practical addition to school supplies. Suitable for students, professionals, and homemakers alike, it helps keep tasks and priorities in check effortlessly.
Fast mode and long-context options
Anthropic lists fast-mode and long-context options as potential premiums. Treat each as a separate line in your estimate and confirm the multiplier for the exact model and feature combination you enable. Do not apply a generic percentage to all requests. Capture the feature flag in usage logs, because a change in the default setting can alter cost without changing the model name.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →US-only inference and partner routing
US-only inference for Claude 4.6 and later adds a 1.1× multiplier according to Anthropic’s pricing documentation. Claude can also be routed through Amazon Bedrock, Google Cloud or Microsoft Foundry. Partner-operated platforms have their own regional pricing and billing mechanics, so keep those figures separate from direct API calculations.
Server-side tools
Tools can add usage charges beyond model tokens. Anthropic lists web search at $10 per 1,000 searches. Count tool invocations and their documented rates as their own line items; a prompt that triggers several searches can cost more than its text-generation component. Record whether a tool call was attempted, completed or retried, and include tool-result text in token counts when it is sent back to the model.
Choosing a model by workload
The cheapest model is not automatically the cheapest complete workflow. Select for the quality and latency your task requires, then compare the total tokens and retries needed to achieve an acceptable result.
- Haiku: positioned for high-volume classification, extraction and routing. Its 5.5 pricing has a major prompt-length step, so keep large documents below the threshold where practical.
- Sonnet: the general speed-and-intelligence balance for many production applications.
- Opus and Fable: aimed at demanding reasoning and long-horizon work, with substantially higher output rates.
A lower per-token rate can lose its advantage if the model needs multiple repair calls, produces verbose answers or fails a structured-output validator. Log successful and failed attempts by model so the decision reflects completed work, not just the price table.
When comparing candidates, use the same axes: input and output price, cache economics, context-window and prompt-length thresholds, latency or fast-mode premium, batch eligibility, rate limits, geography and data-residency premium, billing method, and available operational controls. Keep a fallback model in the forecast, but do not assume it has the same tokenizer, cache rates or retirement timeline.
Controls that prevent runaway usage
Cap generated output
Set a sensible max_tokens for each endpoint. Add schema-level limits for lists, fields and explanations, and reject responses that exceed your product’s useful size. Because output rates are higher than input rates on every model in the snapshot, an unconstrained response can dominate cost.
Stop retry storms
Retry only transient failures, use exponential backoff with a maximum attempt count, and assign a per-request cost budget in the caller. Do not retry validation failures indefinitely: either repair once with a smaller prompt or route the item to a review queue.
Count before expensive sends
Run /v1/messages/count_tokens when a user uploads a document, when conversation history grows, or before a batch is submitted. Reject, summarize or split work that exceeds your approved token budget.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteRank #4
- 【Adjustable & Ergonomic】：This laptop stand can be adjusted to a comfortable height and angle according to your actual needs, letting you fix posture and reduce your neck fatigue, back pain and eye strain. Very comfortable for working in home, office and outdoor.
- 【Sturdy & Protective】 ：Made of sturdy metal, it can support up to 17.6 lbs (8kg) weight on top; With 2 rubber mats on the hook and anti-skid silicone pads on top & bottom, it can secure your laptop in place and maximum protect your device from scratches and sliding. Moreover, smooth edges will never hurt your hands.
- 【Heat Dissipation】 ：The top of the laptop stand is designed with multiple ventilation holes. The open design offers greater ventilation and more airflow to cool your laptop during operation other than it just lays flat on the table.
- 【Portable & Foldable】：The foldable design allows you to easily slip it in your backpack. Ideal for people who travel for business a lot.
- 【Broad Compatibility】：Our desktop book stand is compatible with all laptops from 10-15.6 inches, such as MacBook Air/ Pro, Google Pixelbook, Dell XPS, HP, ASUS, Lenovo ThinkPad, Acer, Chromebook and Microsoft Surface, etc.Be your ideal companion in Home, Office & Outdoor.
Use hard spend ceilings
Configure organization or workspace spend limits. Anthropic says requests return HTTP 400 when a specified spend limit is reached. Treat that response as an operational state: alert the team, stop nonessential jobs and provide a controlled fallback rather than repeatedly resubmitting.
Use rate-limit headers as cost telemetry
Anthropic’s rate-limit documentation says that, for most Claude models, only uncached input tokens count toward the ITPM limit. Cache reads can therefore improve effective throughput while also reducing input cost. Record the response headers alongside usage so a sudden increase in uncached tokens is visible before it becomes a billing surprise.
Instrument every token category
For each response, store model, request ID, input tokens, output tokens, cache_creation_input_tokens and cache_read_input_tokens. Sum the three input categories for total input consumption, but do not assume cache reads consume ITPM in the same way as uncached input. Add tool counts, retry number, workspace and feature flags so finance reports can explain the bill.
Set alerts around behavior, not just dollars
Daily spend is useful, but an output-to-input ratio, cache-hit rate, prompt-length percentile and retry rate often reveal the cause sooner. Alert when output tokens jump while request volume is flat, when cache writes approach cache reads, or when a new feature flag appears in production. Pair alerts with a runbook that can pause background queues without taking down interactive traffic.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →A practical monthly forecasting workflow
- Define workloads. Separate interactive requests, background jobs, evaluations and tool-using flows; they have different latency and batching options.
- Sample real traffic. Measure input and output tokens for each workload rather than estimating from characters or request counts.
- Choose a model scenario. Price the intended model and at least one fallback, including the model’s prompt-length thresholds.
- Apply cache behavior. Estimate write frequency, cache lifetime and reads per write. Keep cached and uncached tokens in separate columns.
- Add operational overhead. Include retries, validation calls, peak-period volume, tools, fast mode, geography and long-context premiums.
- Multiply by volume. Use expected and high-case request counts. Present both so a growth spike does not look like an accounting error.
- Set the ceiling and alert. Configure spend limits, then alert on daily spend, output-to-input ratio, cache-hit rate and retry rate.
- Reconcile. Compare your event log with the Console Usage page and investigate any gap before increasing limits.
Maintain a low, expected and high case. For each case, show requests, average input and output tokens, cache writes and reads, retries, tool calls, model rates and multipliers. This makes a forecast auditable: changing one assumption, such as a 20% increase in output length, immediately shows the cost impact.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Billing, credits and deployment choices
Direct Console billing
Anthropic says Console users generally buy prepaid credits. Successful calls are charged and failed requests are not. Auto-reload is available, and purchased credits expire one year after purchase; the expiration date cannot be extended under the stated credit terms. Set the reload amount and threshold deliberately so a runaway process cannot silently consume an unlimited balance.
Distinguish a failed HTTP request from an unsuccessful application outcome. A failed request is not charged under the stated rule, while a completed API call that returns an answer your validator rejects is still a successful, billable call. That distinction belongs in incident reviews and retry accounting.
Direct API versus cloud marketplaces
The direct API is a REST service at https://api.anthropic.com. For enterprise procurement, Anthropic documents Claude on Amazon Bedrock and Google Cloud, plus Claude in Microsoft Foundry. AWS and Azure marketplace routes convert token-rated usage into Claude Consumption Units (CCUs) at $0.01 per CCU and meter usage hourly; partner regions can apply different pricing. Compare the complete invoice, region, data-residency terms, quotas and observability controls rather than copying direct-API rates into a partner budget.
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 & 11For a procurement comparison, keep separate worksheets for direct Claude API, Claude on Amazon Bedrock, Vertex AI Claude pricing and Claude in Microsoft Foundry. Document currency, taxes, committed-use terms, regional availability and who controls rate limits. A partner route may simplify an enterprise purchase order or residency requirement, but its hourly metering and regional price can make the effective per-token cost different from the first-party table.
Best Value
- Twin-Ball Design : Hiacry gel pen features a double ball conical tip that reduces writing resistance and provides a stable and smooth writing experience. 0.7mm quick drying black ink ensures no jumping, leakage or seepage.
- Quick-Dry Long Lasting Ink：0.7mm black ink can slide effortlessly, dry immediately, & is not easily dirty. It has a high-capacity reservoir (up to 1300 meters), which is very suitable for left-handed & right-handed users.
- Sleek Design for Journaling & Planning: Engineered with a low center of gravity and precision tip for steady ink flow, these journaling pens excel at note-taking, sketching, planning, and more. They combine stylish pens with top-tier performance.
- Rolling Ball Design：The pocket clip with roll ball design can be easily attached to notebook pockets and binders, in addition, the fidget on the pen clip is also a small toy for your daily thinking to relieve stress.
- Comfortable Non-Slip Grip：The ergonomic pen barrel has a soft rubberized coating for a comfortable, non-slip grip, so you won't get fatigued even after long hours of writing.
Common estimation mistakes
- Using one blended token rate: input and output have separate prices, and output is often five times input on the listed models.
- Ignoring prompt thresholds: Haiku 5.5 changes tiers above 100,000 prompt tokens.
- Charging cached tokens twice: cache writes and reads are distinct from uncached input; model each event once.
- Forgetting retries and validators: a failed business-level validation can still create another billable successful API call.
- Assuming a batch discount applies to interactive calls: the 50% reduction is for asynchronous Batch API processing.
- Copying stale prices: model names, retirement status, cache multipliers and rate-limit tiers can change.
- Confusing a failed request with a failed task: failed API requests are not charged under the stated billing rule, but a completed call that returns an unusable answer is still a successful, billable call.
- Mixing deployment invoices: Bedrock, Google Cloud and Azure marketplace charges should not be blended with direct API rates without normalizing region and billing units.
- Budgeting from average prompts only: long-context outliers can cross a pricing tier and consume a disproportionate share of tokens.
Keeping a pricing model current
Record the date you retrieved rates, the model identifier, prompt-length assumptions, cache lifetime, deployment region and enabled features. Re-run token counts when prompts, tool schemas or tokenizers change. Review the official pricing, rate-limit, billing and tool documentation before launch and whenever Anthropic announces a model update. A forecast that exposes its assumptions is easier to correct than a single unexplained monthly number.
Frequently Asked Questions
How does Claude API pricing work?
Anthropic meters input and output tokens separately at model-specific US-dollar-per-million-token rates. Add cache writes and reads, tools, routing multipliers and retries to the token calculation for a complete estimate.
What is the cheapest Claude model?
For prompts up to 100,000 tokens, Claude Haiku 5.5 has the lowest listed rates at $0.10 per million input tokens and $0.50 per million output tokens. Above that prompt length its rates rise to $0.50 and $2.50. The cheapest complete workflow still depends on quality, retries and output volume.
Free tools Windows power users keep installed
One-click scans. No signup required.
Does prompt caching reduce Claude API cost?
It can when a prefix is reused. Five-minute writes cost 1.25 times the input rate and generally pay back after one read; one-hour writes cost twice the input rate and generally pay back after two reads. Unreused prefixes add a write charge without a saving.
How much does the Batch API save?
Anthropic documents a 50% discount on both input and output tokens for asynchronous Message Batches. The discount is intended for deferrable work such as classification, enrichment, backfills and evaluations, not interactive latency-sensitive requests.
Are failed Claude API requests charged?
Anthropic’s billing guidance says successful API calls and completed tasks are charged, while failed requests are not. A completed call that produces an unusable answer is still billable, even if your application later retries it.
How can I calculate a monthly Claude API budget?
Measure input and output tokens by workload, multiply each by the selected model’s rates, then add cache operations, tools, retries, validation calls, feature multipliers and peak volume. Show expected and high-case scenarios and reconcile your event log with the Console Usage page.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Do credits expire?
Console credits are generally prepaid. Purchased credits expire one year after purchase under the stated credit terms, and the expiration date cannot be extended. Auto-reload is available, so set its threshold and amount deliberately.
Does US-only inference cost more?
For Claude 4.6 and later, Anthropic documents a 1.1× multiplier for US-only inference. Include that multiplier only for requests using the feature and keep regional partner-cloud pricing separate.
Can I use Claude pricing from the direct API for Bedrock or Google Cloud?
No. Amazon Bedrock, Google Cloud and Microsoft Foundry have their own regional pricing and billing mechanics. AWS and Azure marketplace routes use Claude Consumption Units at $0.01 per CCU and meter usage hourly, so compare the complete invoice rather than copying first-party rates.
How do I stop Claude API costs from running away?
Set endpoint-specific max_tokens and schema limits, cap retries, count large prompts with /v1/messages/count_tokens, configure organization or workspace spend limits, and alert on spend, output-to-input ratio, cache-hit rate and retry rate. Anthropic says requests return HTTP 400 after a specified spend limit is reached.
The Bottom Line
Estimate Claude API spend from separate input and output token counts, then add cache operations, retries, tools and routing premiums. Use caching for reused prefixes, Batch API for deferrable volume, token counting before large requests, strict output and retry limits, and hard workspace ceilings with usage telemetry. Recheck Anthropic’s current rates and your deployment’s regional terms before relying on any 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.
