Why your Claude Code tokens run out so fast
Every message resends the whole conversation, so one word in a long chat is charged for all of it.
Plays from YouTube when you tap
- File
- DA-C001
- Published
- 6 Oct 2026
- Runtime
- 1:05
- Sources
- 01
- The detail nobody notices
- Message twenty carries all nineteen messages before it.
What’s really happening.
- 01
In a new chat, a one-word message is tiny. At the end of a long chat, that same word arrives with everything you said before it.
- 02
The model keeps no memory between messages, so your app sends the entire conversation again with every request. A one-line question in a chat that has been open all day is charged for the whole conversation.
- 03
Caching softens it: the part that hasn’t changed is charged at a fraction of the price. But the cache expires, and the first message after a long break pays full price for everything.
- Request
- Stateless. Each request carries the full history; Claude Code “sends your full conversation with every request”.
- Growth
- With equal-sized messages, doubling the conversation sends about four times the tokens in total (1 + 2 + … + n).
- Cache
- Unchanged history is re-read at the cached token rate, a fraction of the input price.
- Cache lifetime
- Five minutes by default on an API key, an hour on a subscription. A longer break misses the cache.
- Reset
- /clear between tasks costs nothing. Stale context is paid for on every later message.
The technical layer of this case is behind Agent Mode.
When the task changes, start a new chat. Starting fresh costs nothing.
Transcript
- You sent one word, and it cost you thousands of tokens. Why? And which everyday habit makes it worse?
- Let’s decode it.
- In a new chat, a one-word message is tiny. At the end of a long chat, that same word arrives with everything you said before it.
- Here’s the reason: the model has no memory. So your app sends the entire conversation again with every message. Message twenty carries all nineteen before it.
- That is why a one-line question, in a chat that has been open all day, is charged for the whole conversation. Double the conversation, and you pay about four times as much.
- Caching softens it. The part that hasn’t changed is charged at a fraction of the price. But the cache expires. Step away for too long, and your next message pays full price for everything.
- So when the task changes, start a new chat. Starting fresh costs nothing.
- You now know why your tokens run out. Join the Decode Agent journey, and decode the next case with me.
No corrections since publication. If a fact changes, it’s logged here.
You already see what others miss_
Claim your Agent ID and follow the journey from the inside. Decode Agent is just getting started.
- Your Agent ID, yours to keep and share.
- Every case, filed here with its sources.
- On the list when updates begin.
