Fieldbook to AIThe plain-English guide to which AI to use

Best of

Cheapest good AI models

For lots of everyday work at the lowest cost, Claude Haiku 5.5, GPT-6 Luna and DeepSeek V4.1 Flash are the main low-cost choices; Schematron is the cheapest for turning web pages into data.

Last updated .

Costs here use the prices AI companies charge software developers, so they show relative cost, not your monthly bill. See what ChatGPT, Claude and Gemini charge you.

Model and our take (editorial)TypeSmartnessReads at once*EssayChatPDF
Schematron V2 Turbo
Made by Inference.net.A tiny model that does one thing: turn a web page into a neat table of data, very cheaply.
Single-purposefree to downloadNot rated yet190 pages<1¢$0.04<1¢
MiMo-V2.6-Flash
Made by Xiaomi.One of the cheapest models we track. It understands audio and video too, and can be downloaded and run yourself.
Everyday and cheapfree to download★★★☆☆1,580 pages<1¢$0.18<1¢
DeepSeek V4.1 Flash
Made by DeepSeek.Very low price, can be downloaded and run yourself, and reads a very large amount at once. Hard to beat for big batches of work.
Everyday and cheapfree to download★★★☆☆1,570 pages<1¢$0.19$0.01
Qwen3.8 Flash
Made by Alibaba (Qwen).A cheap model that understands pictures and video, and can be downloaded and run yourself.
Everyday and cheapfree to downloadNot rated yet1,500 pages<1¢$0.19$0.01
Claude Haiku 5.5
Made by Anthropic, the company behind Claude.The cheap and fast Claude. Great for summaries, quick questions and anything you do a lot of.
Everyday and cheap★★★★☆1,500 pages<1¢$0.13<1¢
GPT-6 Luna
Made by OpenAI, the company behind ChatGPT.OpenAI's fast, cheap model for chat, sorting things into groups and light tasks.
Everyday and cheap★★★☆☆1,580 pages<1¢$0.13<1¢

Essay = a 1,000-word essay; Chat = a 100-message conversation; PDF = summarizing a 50-page PDF. *A rule of thumb: one page is about 500 words, and a model "reads" about 0.75 words per token. Smartness is our 1 to 5 banding of the Artificial Analysis Intelligence Index (how); "Not rated yet" means no score is available. Costs are computed from list prices using the assumptions on How we work.

How we chose: Editorial shortlist. Test on a small sample of your own work before relying on a cheap model.

Why each one

Schematron V2 Turbo
A tiny model that does one thing: turn a web page into a neat table of data, very cheaply.
MiMo-V2.6-Flash
One of the cheapest models we track. It understands audio and video too, and can be downloaded and run yourself.
DeepSeek V4.1 Flash
Very low price, can be downloaded and run yourself, and reads a very large amount at once. Hard to beat for big batches of work.
Qwen3.8 Flash
A cheap model that understands pictures and video, and can be downloaded and run yourself.
Claude Haiku 5.5
The cheap and fast Claude. Great for summaries, quick questions and anything you do a lot of.
GPT-6 Luna
OpenAI's fast, cheap model for chat, sorting things into groups and light tasks.