Agent token savings often come from context control, not model choice
AI agent may not be mainly solved by switching to a cheaper model. In the examples collected here, bigger savings came from limiting what the agent is allowed to put into context. Large, unpredictable outputs can consume many tokens quickly; one Codex example cut token use by about half by adding one rule that capped shell output when the size was unknown.
were another major cost. One team had 508 and was spending about $377 per run because those were sent again on every call; by not sending every tool description upfront, the cost dropped to about $29 per run. Another measurement found that about 67,000 tokens were already used before the person asked the first question.
Choosing smaller or cheaper models can help, but the larger savings usually come from controlling tool output, tool lists, and the starting context.
Key points
- Switching to a cheaper model can help, but it may not be the biggest source of savings.
- Large outputs can quickly fill the context and waste tokens.
- A Codex example cut token use by about half with an rule that limited unknown-size shell output.
- A team sending 508 MCP on every run cut cost from about $377 to about $29 per run.
- One measurement found about 67,000 tokens used before the first user question.