Guides
Practical writing on LLM API pricing, grounded in the price data we record every day. Maintained by Jaehyun Jeon.
How to Estimate Your LLM API Bill Before You Ship
A practical, worked method for turning token counts into a monthly LLM API bill — input vs output, caching, and the mistakes that blow up the estimate. Uses live pricing from computetrail.
Why Output Tokens Cost 5–6× More Than Input
Across the major LLM APIs, generated (output) tokens cost five to six times more than prompt (input) tokens. Here's why that ratio exists and how it should change what you build.
GPT vs Claude vs Gemini: An API Pricing Teardown
A tier-by-tier price comparison of OpenAI, Anthropic, and Google's LLM APIs — flagship, mid, and budget — with blended costs, context windows, and cache rates from live data.
Are LLM API Prices Actually Falling? What Daily Tracking Shows
Two weeks of daily price recording across 324 models shows the 'prices always fall' story is too simple — most prices don't move, and when they do, they move both ways.
Prompt Caching Quietly Cuts Input Cost ~90% — Here's the Math
Most production prompts repeat a big fixed chunk on every call. Prompt caching bills that chunk at a fraction of the input price — around 90% off on major models. A worked breakdown of when it pays and when it doesn't.
Which LLM Prices Actually Moved in 70 Days of Daily Snapshots
Seventy days of once-a-day price recording across 361 models: three quarters never moved, the ones that did moved up more often than down, and the changes cluster on a handful of dates.
Batch APIs: Is It Really Half Price?
We compared all 69 batch variants against their standard counterparts in one day's snapshot. 55 are exactly 50% off. The other 14 are the interesting ones — including several that cost more.
Context Window vs Price: What the Data Shows
Across 321 paid models, a bigger context window barely predicts a higher price — rank correlation just +0.26. Inside a single size bracket, prices span 20,000×.
LLM API Price Report: August 2026
OpenAI moved six models in opposite directions on a single day, three providers quietly ended their batch discount, and the market median finished the month exactly where it started.
LLM API Price Report: July 2026
31 days of daily snapshots: 20% of tracked models repriced, increases outnumbered cuts 38 to 10, and OpenAI's new GPT-5.6 Luna lost 90% of its price three weeks after launch.
Figures in these guides come from computetrail's daily snapshots — seemethodology. Prices move; each guide notes the snapshot it used.