What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
To lower Claude API costs, reuse stable prompt content with prompt caching, remove input the task does not need, and compare actual input and output usage with current prices. Caching can make repeated input cheaper, but it does not eliminate the initial cache-write charge or guarantee a hit. The savings depend on how often the same prefix is reused, how much of each request changes, and which model and cache duration you use.
How prompt caching changes API costs
Anthropic prompt caching reuses previously processed prompt content across API calls when a request matches a recent cached prefix. A cache miss processes the prefix and writes it to cache; a hit reuses it. The initial write costs more than ordinary input, while cache reads generally cost less. New or changing content still incurs its applicable input charge.
As an Amazon Associate I earn from qualifying purchases.
Good candidates are content that is both substantial and genuinely repeated: system instructions, examples, tool definitions, reference documents, or recurring conversation context. Content that changes on every request is usually a poor candidate for the stable prefix.
Current cache durations and price multipliers
Anthropic’s prompt caching guide documents a 5-minute default time-to-live (TTL) and a 1-hour option. Its pricing page, accessed October 7, 2026, lists the following multipliers against the model’s base input price; model-specific exceptions apply to cache reads.
#1 Best Overall
| Cache operation | Documented rate | What it means |
|---|---|---|
| 5-minute cache write | 1.25× base input price | Writing the prefix costs more than ordinary input. |
| 1-hour cache write | 2× base input price | The longer-duration write has a higher premium. |
| Cache read | Generally 0.1× base input price | Exceptions listed on the pricing page include Claude Fable 5.1 and Mythos 5.1 at 0.025×, and Opus 5.5 at 0.05×. |
Anthropic says the listed pricing breaks even after one cache read for the 5-minute duration and two for the 1-hour duration. Treat that as a rule of thumb for comparable cached tokens, not a savings guarantee: misses, reuse timing, changing content, and model-specific rates affect the result. The TTL runs from the start of the request that writes or reads the entry, and response generation time counts against it. A long response can use up much of a 5-minute window before the next call.
Rates and available models can change. For a concrete illustration of the pricing page accessed October 7, 2026, Claude Sonnet 5.5 was listed at $2 per million input tokens and $10 per million output tokens; 5-minute cache writes were $2.50 per million and cache reads $0.20 per million. Check Anthropic’s current pricing page for the model you actually use and compare input, output, cache-write, and cache-read rates.
Choose what to cache and where the breakpoint goes
A cache hit depends on matching the prompt prefix through the cache breakpoint. Put stable material before the breakpoint and request-specific content after it. Marking a prompt does not guarantee a hit: the prefix must meet the applicable model’s minimum cacheable length, and requests below that minimum are processed without caching.
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 →Automatic caching
Anthropic describes automatic caching as a straightforward starting point for many cases. It uses a top-level cache_control field and manages the breakpoint as a conversation grows. This can suit a stable conversation or prompt structure where you do not need to choose each cached section yourself.
Explicit breakpoints
For finer control, attach cache_control to selected blocks. Anthropic says one breakpoint at the end of stable content is sufficient in most cases. Multiple breakpoints can help when sections change at different rates or a long conversation moves beyond the cache lookback; up to four are supported. Consult the prompt caching guide for the current request format and model-specific requirements.
Shorten prompts without weakening the task
Remove repeated, stale, or irrelevant context rather than stripping out instructions the model needs. Avoid repeating the same examples or directions across turns when they can be supplied once in a reusable prefix. Keep the remaining instructions explicit and relevant; an overly compressed prompt can make the requested result ambiguous.
- Delete reference material that does not affect the current request.
- Remove duplicate instructions and examples, especially when they are already present in stable context.
- Keep task-specific details, constraints, and output requirements needed for a correct response.
- Evaluate prompt changes on representative tasks. If a shorter prompt leads to inadequate answers or extra retries, the workflow may cost more overall.
Anthropic’s token-counting endpoint estimates input size for structured message inputs before you send a request. Use it to compare candidate prompt versions, budget tokens, or inform model routing. The estimate can differ slightly from actual message usage, and token counting does not simulate prompt caching. See the token counting documentation.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallMeasure real costs and confirm cache activity
Use response usage fields to distinguish cache activity from uncached input. Anthropic defines total input as the sum of the three fields below; when caching is in use, input_tokens alone is not the full prompt input.
Best Value
cache_creation_input_tokens: input tokens written to a cache entry.cache_read_input_tokens: input tokens retrieved from cache.input_tokens: tokens after the last cache breakpoint that were not read from or written to cache.
Token counting estimates a prompt before message creation; it does not verify cache hits. Confirm caching from actual message-response usage data, then compare representative traffic, including cache writes, reads, misses, uncached input, and output tokens. Apply the current rate for each category and model. Anthropic’s prompt caching guide and pricing documentation describe the usage fields and billing categories.
A practical cost-reduction workflow
- Identify repetition: Find large, stable prompt sections reused across calls. Do not cache material merely because it is long.
- Separate stable from changing content: Place reusable context before the breakpoint and per-request details after it.
- Pick a caching approach and duration: Start with automatic caching for simpler management, or use explicit breakpoints when sections change independently. Choose between 5 minutes and 1 hour based on actual reuse intervals and the higher write premium for the longer TTL.
- Trim unnecessary input: Remove stale or duplicated material while retaining the instructions needed for reliable output.
- Estimate and measure: Use token counting to compare prompt variants, then inspect actual response usage to verify cache reads and total input.
- Compare complete request costs: Apply current model-specific input, output, cache-write, and cache-read rates to representative requests. Include output tokens and any retries when assessing whether a change helps.
These changes can reduce repeated input charges, but there is no dependable universal savings percentage: workloads differ in reuse frequency, prompt size, cache hits, output volume, and model price. Anthropic also notes that clearer, specific instructions help Claude understand the requested output; cost reduction should not come at the expense of task quality.
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.
Recommended Free Tools
