Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Anthropic does not charge a fixed $30 for a Claude inference. Its pricing is per token, set by model and by whether the tokens are input or output, with extra charges for some server-side tools. A $30 bill from an agent is an accumulated total: dozens of model turns, each re-reading a growing context, plus tool results and any metered tools. This article shows how that total builds up, with a worked example, and what to change to bring it down.
What the listed prices say
Anthropic’s Claude Platform pricing documentation (page accessed October 5, 2026) lists these rates. Prices change, so confirm them on the live page before budgeting.
As an Amazon Associate I earn from qualifying purchases.
| Model | Input, per million tokens | Output, per million tokens |
|---|---|---|
| Claude Opus 4.7 | $5 | $25 |
| Claude Sonnet 5 | $2 | $10 |
Web search is billed separately. In Anthropic’s words: “Web search is available on the Claude API for $10 per 1,000 searches, plus standard token costs for search-generated content.” Tool definitions, tool-use blocks and tool results all count as tokens too.
Recommended Free Tools
Anthropic’s Sonnet 5 announcement, updated August 10, 2026, says the initial $2/$10 pricing became permanent. It also notes that the newer tokenizer can produce more tokens for the same text, depending on content. A token count from an older model therefore may not transfer directly to a new one.
#1 Best Overall
Why an agent loop is not one inference
Anthropic’s engineering post on advanced tool use puts it plainly: “Each tool call requires a full model inference pass.” An agent that searches, reads, reasons, calls another tool and repeats is making many billed requests. Three things inflate the total:
- Repeated turns. Each round trip is a separate request that carries the conversation so far, so input tokens are billed again on every turn.
- Context pollution. Large intermediate results, such as a full web page or a big query response, stay in context and are re-billed on every later turn. Anthropic names this, along with repeated inference, as a cost and latency driver.
- Metered tools. Server-side tools such as web search add their own charge on top of the tokens.
A worked example of how $30 happens
The numbers below are a hypothetical workload I constructed to show the arithmetic at the listed rates. They are not a measurement. Assume a 60-turn agent run with no prompt caching, an average of 90,000 input tokens per turn, and 2,000 output tokens per turn.
| Line item | Opus 4.7 ($5 / $25) | Sonnet 5 ($2 / $10) |
|---|---|---|
| Input: 60 × 90,000 = 5.4M tokens | $27.00 | $10.80 |
| Output: 60 × 2,000 = 120,000 tokens | $3.00 | $1.20 |
| Total | $30.00 | $12.00 |
Input dominates, because the context is re-sent each turn. That is why trimming context usually saves more than trimming the model’s replies. Sonnet 5 is cheaper here, but whether it completes the task as reliably is something you have to test on your own workload. Price per token says nothing about success rate.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Build a cost model from your own usage records
Don’t multiply a vague count of “inferences” by a flat figure. Read the usage fields returned with each request, and break each run down into:
- uncached input tokens;
- cache writes and cache reads (check the pricing page for the current multipliers and cache duration);
- output tokens;
- tool-call overhead: definitions, tool-use blocks and results;
- server-side tool charges, such as per-search fees;
- any platform- or region-specific charges on your billing route.
Then compare options along model price, input/output mix, cache hit rate, number of round trips, size of tool results returned to context, and metered tool usage. If you quote “$30 per task” internally, record the model, token counts, cache mix, tool usage, billing platform, geography and date next to it.
Ways to cut the bill
Keep intermediate data out of the context
Anthropic’s Programmatic Tool Calling lets a script process intermediate tool results and return only the final output to Claude. On its own complex research tasks, Anthropic reports average usage fell from 43,588 to 27,297 tokens, a 37% reduction. That is Anthropic’s result for those tasks, not a guaranteed saving elsewhere. It also cuts round trips, since the script can chain tool calls without a model pass for each.
Shrink tool results
Return only the fields the agent needs. Truncate or summarise large pages and query results before they enter the conversation, since every retained token is re-billed on each later turn.
Cache the stable prefix
System prompts and tool definitions repeat on every turn, which makes them natural candidates for caching. Your cache hit rate then becomes a metric worth tracking.
Match the model to the step
Compare models on representative tasks, including reasoning effort and tool use. Anthropic’s announcement characterises cost-performance as dependent on the task and the effort level. Measure cost per successful task, not cost per token.
Cap the loop
Set a maximum turn count and a per-run token budget, and alert on outliers. A stuck agent retrying the same failing tool call is the most common way for one run to cost many times the average.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Enterprise visibility
An Anthropic event listing dated September 15, 2026 describes enterprise controls: model defaults and entitlements, per-teammate spend visibility, natural-language cost questions through Analytics Chat, and usage and cost reporting through the Analytics API. The listing advertises these features but does not quantify any savings, so check availability for your plan.
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.
