LLM news, with our take
Headlines from across the AI world, each with a short note from us on what it means for your costs. We link to the source and write the takes ourselves.
Site updates
Eight providers, sliders, and a planning calculator
June 11, 2026
The cost comparison now covers 26 models from eight providers: Anthropic, OpenAI, Google, xAI, DeepSeek, Mistral, Amazon Nova, and Cohere. You can drag sliders to size a request instead of typing numbers, filter the table by provider, and sort any column.
We also added a calculator that answers a question we kept asking ourselves: is it cheaper to plan one good request, or to fire off a string of follow-ups? Try it on the home page under "Plan vs. piecemeal."
One note on the numbers. We verify Anthropic's prices against the official reference. The other providers change prices often and the public pages do not always agree, so we mark those as indicative. Check the provider before you budget around them.
Why one good request usually beats ten follow-ups
June 11, 2026
Every time you send a follow-up, the model has to read the whole conversation again. Turn five pays for turns one through four all over again, as input tokens. Stack up ten short questions and you can pay several times what one well-planned request would have cost, for the same work.
There is a real exception: prompt caching. When the repeated part of your prompt is cached, you pay roughly a tenth of the normal input price for it. That closes most of the gap. So the practical advice is simple. Say what you need up front in one request when you can. If you do go back and forth, turn caching on.
The home-page calculator lets you put in your own numbers and see both paths side by side.