← All notes

Where context compression actually saves agent tokens

Short answer

Measure each lever separately. In one gateway study, trimming tool schemas was the only reliably positive saving per turn; compressing file reads paid off over many turns. Another method cut cost up to 50% by only truncating or dropping content, never rewriting it. [S1][S2]

Tips

Claims

Sources

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted).

License: CC-BY-4.0 · Summary by iinventedlight (AI-assisted). Sources: arXiv, [2609.22114] An Empirical Cost Attribution of Context-Compression Gateways in Multi-Turn Coding Agents, CC BY 4.0; arXiv, [2609.26779] CliffCompaction: Cost-Efficient Compaction for Long-Horizon Coding Agents, CC BY 4.0.

Cite this

Permanent ID: iil:note:swmmb7pk

iinventedlight. 2026. "Where context compression actually saves agent tokens." iinventedlight Library, iil:note:swmmb7pk, version of 2026-11-04. https://iinventedlight.com/library/notes/ao-cost-compression-accounting/. License: CC-BY-4.0.

@misc{iil_note_swmmb7pk,
  author = {{iinventedlight}},
  title = {{Where context compression actually saves agent tokens}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-04},
  url = {https://iinventedlight.com/library/notes/ao-cost-compression-accounting/},
  note = {Permanent ID iil:note:swmmb7pk. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt. This page in Markdown: /library/notes/ao-cost-compression-accounting.md.