Agent harness token overhead: where it goes
Two harnesses on one model can bill an order of magnitude apart. The mechanics behind the gap, and how to measure your own agent's token spend.
#tokens
Filtering by topic #tokens · clear
Claude Pro in Portuguese: limits and tokens
Claude Pro spends your limit in tokens rather than messages, and Portuguese costs more tokens than English. Measure the gap and make your quota last.
Autoprompt skill: fewer failures, more tokens
Autoprompt solves 73 of 89 Terminal-Bench 2.1 tasks against a baseline of 60, at roughly 3x the time and 2x the tokens. What that costs on a rented VPS.
Reasoning effort settings on a local LLM
Reasoning effort sets how long a local model thinks before it answers. What the levels change on your own hardware, and how to measure the real cost.
Ollama num_predict: cap the output length
num_predict caps how many tokens Ollama will write. The three places to set it, which one wins, and how to read done_reason in the response.
Do Claude's memory features cost extra?
Anthropic publishes no memory token price. Remembered text is replayed as input tokens, so your cost tracks how much you replay and whether it is cached.
Agent skills vs MCP servers vs rules files
Three ways to give a coding agent context, and one rule for picking. See what a skill, an MCP server and a rules file each cost in tokens and upkeep.
Paritok token gateway: cheaper agent bills?
Paritok compresses the file reads and tool output your coding agent sends. The project claims 74% fewer tokens. Here is the mechanism and the break-even math.
Claude Code spend tracking tools compared
Claude Code spend trackers answer different questions. Compare local log parsers against the built-in usage screens and your own OpenTelemetry stack.
Input vs Output Tokens: Claude's Cost Gap
Claude output tokens cost five times input. Here is why decoding is slower than prefill, and what that asymmetry does to a real agent bill each month.
Claude Fable 5: price and when to use it
Claude Fable 5 lists at $10 per million input tokens and $50 per million output. Here is what the published rates mean for the jobs you actually run.
Why is Claude so expensive? The token math
Output tokens cost five times input, and every turn re-reads your whole context. See the arithmetic on one real session, and the four levers that cut it.
How much is 1M tokens in Claude?
Claude bills per million tokens, input and output priced separately. Here is the arithmetic that turns a token count into a number on your monthly bill.
Claude vs ChatGPT API: Which Is Cheaper?
Neither API wins every workload. Learn the token arithmetic once, then read three costed jobs and see exactly where the Claude bill runs higher.
Claude Code context: stop slow, costly sessions
Every turn re-sends your whole context, so a long session slows down and costs more. Read /context, cut the fixed tax, then clear and compact with intent.
What are tokens in Claude? Code session costs
A Claude token is about 3.5 characters of text. Here is why one Claude Code turn bills 80,000 of them, and why five idle minutes make the next turn cost 5x.