Token test: pretty-printed JSON costs 3x CSV
10/03/2026 — 10/03, 01:50·1 sources·1 reports
Story overview
On October 3, 2026, a post on DEV Community described a small hands-on test in which the author wrote the same dataset out in seven different formats and counted the tokens each one cost.
The data was a product table with 20 rows and five fields: id, name, price, in-stock status, and category. Using the o200k_base tokenizer, the author counted tokens for every format and found that indented JSON consumed roughly three times as many tokens as CSV. According to the post, pasting the output of JSON.stringify(data, null, 2) straight into a prompt is the most token-hungry of the common approaches — and the author notes he had never previously measured what that habit actually costs.
The report stops at that single personal comparison. It does not give the token counts for the other six formats, and it does not include any response from other organizations or products. No broader claims about causes or downstream effects appear in the source material.
AI-generated from 1 reports · updated 1 hour ago
Latest turnThe author took one product table of 20 rows and 5 fields, wrote it in seven formats, and counted tokens with the o200k_base tokenizer. Indented JSON came out roughly 3x more expensive than CSV, making JSON.stringify(data, null, 2) one of the costliest habits when pasting data into prompts.
Reports on this story headlines open the original
The author took one product table of 20 rows and 5 fields, wrote it in seven formats, and counted tokens with the o200k_base tokenizer. Indented JSON came out roughly 3x more expensive than CSV, making JSON.stringify(data, null, 2) one of the costliest habits when pasting data into prompts.
DEV Community · AIAI score 80
Other stories people are talking about
- 625SurgeNVIDIA launches 64GB DGX Spark desktop AI computer at $4,9997 sources
- 495RisingGoogle releases Gemini 4 Argon as next-gen frontier AI model14 sources
- 389NewApple tightens macOS Full Disk Access over AI agent risks4 sources
- 224SurgeMicrosoft AI releases MAI-Transcribe-2-Streaming real-time transcription model3 sources
- 191NewAnthropic launches Claude Frontier Academy with $100M2 sources
- 179NewHugging Face open-sources AstaBrief for fast report generation2 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion