---
title: 'In the News: August 14, 2026, Evening'
description: 'Anthropic published the cost model behind a Claude Code session, and the prompt cache expires after an hour, which prices unattended work above supervised work.'
canonical_url: 'https://darkfactory.dev/news/2026-08-14-evening'
markdown_url: 'https://darkfactory.dev/news/2026-08-14-evening.md'
collection: news
date_published: '2026-08-14T20:35:00-04:00'
date_modified: '2026-08-14T20:35:00-04:00'
---

# In the News: August 14, 2026, Evening


Anthropic published the mechanics of what a Claude Code session costs, and one of the
stated numbers is a problem for anyone running agents unattended: the prompt cache expires
after an hour on a subscription, five minutes on an API key. Walk away, come back, and the
whole conversation prefills again at full price. The same vendor's changelog, two days
running, went the other way and made forked subagents inherit the cache outright.

## 1. Anthropic priced the session, and idle time is the expensive part

**[Maximizing the value of your Claude Code sessions](https://claude.com/blog/maximizing-the-value-of-your-claude-code-sessions)** · Lydia Hallie, Anthropic · claude.com, August 14, 2026

A first-party account of what a Claude Code turn is billed for and what breaks the prompt
cache. The figures are Anthropic's own: output tokens are priced "at roughly 5x input",
reading from the cache costs 0.1x the input price, and writing into it costs "up to 2x".
Every turn re-sends the entire conversation to that point, so the working bill is cache
reads on the history, full input price on whatever is new, and output price on the response.

Cache invalidation depends on the request prefix. Requests always go out in
the same order, tool definitions, then system prompt, then the conversation with `CLAUDE.md`
at the front, and the cache "has to match from the very start of the request forward".
Anything that changes the prefix throws the rest away. `/model` and `/effort` are both part
of what the cache is keyed on, so switching either mid-conversation prefills the whole
conversation again at full price. Fast mode is keyed on too, and the re-prefill happens at
fast mode prices. `/compact` replaces the conversation, so nothing behind the system prompt
matches afterwards, which is why the post says compacting is "a lot cheaper before a long
break than after one". And `/rewind` is offered as the cheap alternative when the last few
turns went nowhere, because it cuts turns off the end and leaves everything in front of them
cached.

Command output over 30,000 characters is written to a file with only a preview and path left
in the conversation
(`BASH_MAX_OUTPUT_LENGTH`), so the expensive case is the noisy command that comes in just
under the limit and then rides along on every remaining turn. And a `/loop` "fires as a full
turn in the session you set it up in, carrying that whole conversation with it every time",
with a cache miss on top if more than an hour has passed.

One limitation, stated in the post itself: the two cost curves illustrating model choice
carry the note "Curves are for illustration purposes only. They do not represent real
benchmark data." Nothing above is drawn from them.

**Why it matters:** For the first time, Anthropic has laid out its cost model at enough resolution to expose
a lights-out penalty. Every mechanism that makes a session cheap assumes an operator at the
keyboard: compact before you step away, rewind rather than compact, keep turns close enough
together that the cache is still warm. An hour of idle is a full re-prefill. Unattended
operation is structurally more expensive per unit of work than supervised operation under
this pricing, and that penalty comes from the billing model
rather than from any limit on what the agent can do.

## 2. Forked subagents now inherit the prompt cache, and the todo tools are gone on newer models

**[Claude Code changelog, 2.1.232 and 2.1.233](https://code.claude.com/docs/en/changelog)** · Anthropic · code.claude.com, August 13 and 14, 2026

Two releases landed after this feed last covered the changelog at 2.1.224. In 2.1.232,
"subagent forking is now on by default", and a forked subagent "inherits the full
conversation and prompt cache". Non-teammate agent spawns in interactive sessions now run in
the background by default. The same release adds `/config` rows for cross-session inbound
message handling, GitLab token redaction across nine prefix families, and GitLab plugin
marketplaces.

2.1.233, dated today, removes something people build on: "Todo/task-tracking tools
(TaskCreate/Get/Update/List, TodoWrite) are no longer available on Opus 4.8, Sonnet 5, Fable
5, Mythos 5, and newer models", restorable with `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`. It also
adds an opt-in memory cgroup for Bash commands on Linux (`CLAUDE_CODE_TOOL_MEMORY_LIMIT`) so
"a runaway build can't stall the session", and a configurable WebFetch URL cache TTL
(`CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS`, default unchanged at 15 minutes).

**Why it matters:** Cache inheritance on fork is the direct counterweight to item 1. Until
now a subagent paid for its own prefix from scratch, which is why the guidance has been to
use one only when the job produces output you do not want to keep. A fork that arrives with
the conversation and the cache already paid for changes that arithmetic, and it is on by
default, so it changes for people who did not opt in. The todo-tool removal cuts the other
way: if your harness or your prompts assume `TodoWrite` exists on a current model, they
stopped working on upgrade, and the environment variable is the migration path rather than
the fix.