prefixlens: run.trace.jsonl

5 model calls · 1 lane · 5 full-request, 0 usage-only · generated 2026-11-17 08:33 UTC

1 miss cost $0.24 at list prices; 1 from tool-list changes

Cache hit rate
59.6%
147k of 246k prompt tokens read from cache
Cost at list prices
$0.54
$0.50 on uncached input and cache writes; API list prices, not what was paid
Lost to misses
$0.24
1 call missed cached tokens unexpectedly
Calls
5
1 cold · 3 hit · 1 missed

Costs are at API list prices: what these calls would cost billed per token, not what was paid. Calls on a subscription (a Claude plan, ChatGPT sign-in, Codex) are not billed per token, and discounts, batch and priority tiers are not applied.

prices from the bundled pricing.json (as of 2026-11-12)

Prompt tokens per call

turn 1: 48,212 prompt tokensturn 2: 48,632 prompt tokensturn 3: 49,012 prompt tokensturn 4: 49,912 prompt tokensturn 5: 50,262 prompt tokens

read from cache written to cache uncached input miss intentional break

Misses by cause

causemissescost
tool-list changes1$0.24

Misses

turn 4: miss, tool list changed (+1 tool: 3 → 4), ~$0.24

lost 49,000 of 49,000 cached tokens · vs turn 3 · confirmed · first changed block: #4 of the request (tools)

+ web_search

All calls

turntime (UTC)modelpromptreadwriteuncachedouthitcoststatus
110:00:02claude-opus-5-548,212048,200121400%$0.24cold
210:00:09claude-opus-5-548,63248,200420129599.1%$0.01hit
310:00:17claude-opus-5-549,01248,6203801211099.2%$0.01hit
410:00:26claude-opus-5-549,912049,900121600%$0.25miss: tool list changed (+1 tool: 3 → 4)
510:00:34claude-opus-5-550,26249,900350126099.3%$0.01hit