惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
MongoDB | Blog
MongoDB | Blog
博客园_首页
博客园 - 三生石上(FineUI控件)
博客园 - 聂微东
B
Blog RSS Feed
D
Docker
IT之家
IT之家
大猫的无限游戏
大猫的无限游戏
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
阮一峰的网络日志
阮一峰的网络日志
罗磊的独立博客
Recent Announcements
Recent Announcements
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
A
About on SuperTechFans
The GitHub Blog
The GitHub Blog
G
Google Developers Blog
V
V2EX
量子位
雷峰网
雷峰网
月光博客
月光博客
云风的 BLOG
云风的 BLOG
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
T
Tailwind CSS Blog

HN's home page

Rainbow Query Language | Hacker News Exec into Node via Kubectl An AI native hedge fund The Seven-Action Documentation Model | Hacker News Package Manager for Kubectl Plugins Tongan Castaways | Hacker News Tech overlords plan for conscious AI to conquer the cosmos. What could go wrong? Data Breach Disclosure Lag Is Getting Worse How LLMs Work | Hacker News I Dropped PRDs for Shape Up Go Experiments Explained | Hacker News FCA's Palantir deal could expose UK financial data to Trump's US, critics fear WebXR BCI for Neural-Adaptive Avatar Control in Mixed Reality The first murder conviction via DNA analysis Tom Interviews Theo de Raadt of the OpenBSD Project (2019) [video] Show HN: Replace shell commands with bun shell typescript scripts Quay.io Is Down | Hacker News AI driven analysis of brokerage account fees in the UK Bill Gates Spent Years Crafting His Image. Now It's Cracking Using LLMs to secure source code Wi-Fi 8 in the Lab [video] The household battery revolution that could change energy bills and the world Is Python Becoming Pinyin? | Hacker News Livia – Executive Assistant | Hacker News FindMyPipe – Query Apple Find My from Linux for AI Agents Show HN: Agent skill for creating product launch videos with Remotion RecruitMyself – AI job search copilot for resumes and applications AI coding agents and the erosion of system understanding The 'Resting' Generation and South Korea's Youth Recession AMD Computex 2026: 10 Years of AM4, AM5 Support Through 2029
Changes that cut our LLM pipeline costs more than model-s...
Abbas_Maka · 2026-06-20 · via HN's home page

I have been building multiple LLM systems and for our Organization biggest cost savings weren't from prompt-wordsmithing or model switchings. Sharing useful to anyone watching their token bill :

1) JSON → TOON for structured output: JSON was not made for LLMs. well you can implement your own verison that fits for your needs that reduce tokens usage but what worked for us was TOON. TOON cut output our tokens by ~30% same information, way less syntax tax.

2) Full markdown/HTML → condensed markdown: Using markdown for writing your prompts, getting intermediate results or communication between your Agents eats a lot of tokens. We swithced to condesed markdown and short system prompts that replicate Caveman. this alone cut just on input token costs ~50% on calls that pass prior context forward which can be implemented between Agent Calls.

3) Long Do/Don't instruction lists → 2-3 multi-shot examples: Counterintuitive one - replacing a large lists of DO's and Don'ts for agents rules don't help. rather couple of concrete examples that convers major and all cases actually improved output quality more reliably and it's usually fewer tokens once the instruction list gets long enough to cover real edge cases.

I have seen most people on this sub reddit talk about using open-source or cheaper models. Like we were spending thousands of dollar's but this all changes alone helped reduce cost by 60%.

edit: Open to Discussion, anyone whether something similar would help their setup.