Cheaper tokens did not make agents cheap, they made the meter run longer. agent tasks now burn 5 to…
cheaper tokens did not make agents cheap, they made the meter run longer. agent tasks now burn 5 to 30 times the tokens of a single prompt, and code workloads can pass 1000 times, so a task that costs pennies as one call becomes pounds as a loop.
the price war is winning the wrong metric.
Context
The arXiv paper How Do AI Agents Spend Your Money? (2604.22750) analyzes trajectories from eight frontier LLMs on SWE-bench Verified using the OpenHands framework. It states agentic tasks consume about 1000x more tokens than code reasoning and code chat, and that input tokens dominate cost even with caching. A secondary vendor blog of 16 July 2026 attributes 5 to 30 times to Gartner via NeuralWired.
The 1000x is first-party but limited to coding agents on SWE-bench Verified against code reasoning and code chat, not a single prompt in general. The 5 to 30 times is secondary and second-hand, and the Gartner primary source was not read, so it is unverified as a first-party figure. A 3500x figure versus a single-round reasoning baseline appears in the paper but its full wording was cut off in the fetch. Pennies to pounds has no source, and the note uses pounds while the sources use dollars. The price war is winning the wrong metric is the author's opinion.
Related work
- How are AI agents spending your tokens (Stanford Digital Economy Lab) ↗Summary of the paper, snippet only.
- Task-specification effects on token spend (arXiv 2608.25399) ↗Snippet only.
Watch next
- The primary Gartner document and the paper's per-model cost tables.
Sources
- How Do AI Agents Spend Your Money? (arXiv 2604.22750)arxiv.org
- Agentic AI inference cost 2026 (Spheron)spheron.network
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 18 September 2026 at 18:17 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →