Loading...
Most affordable AI models ranked by cost per million tokens
GPT-4.1 Nano is the cheapest credible option at $0.1/$0.4, undercutting everything else by a wide margin, and it suits high-volume, simple text tasks where capability ceilings barely matter.
Rankings based on public benchmark data. Prices in USD per 1M tokens (direct provider). Updated June 2026.
$0.1 input, $0.4 output. Nothing else in this lineup comes close. GPT-4.1 Nano is built for the work where cost per token is the only number that matters and the task is simple enough that a small model handles it fine.
The trap is treating cheapest as best. Nano is a lightweight model. For classification, tagging, short summaries, and bulk text transformation, it does the job and the savings at scale are enormous. For anything requiring real reasoning or careful code, it will disappoint, and you will pay more fixing its output than you saved generating it.
The near-budget tier deserves a look before you commit. DeepSeek V3 runs $0.27/$1.1 and codes at 50 — far more capable for a small step up. Gemini Flash Lite at $0.25/$1.5 and GPT-4.1 Mini at $0.4/$1.6 sit in similar territory, each buying meaningful capability over Nano. DeepSeek R1 deserves mention too: at $0.28/$0.42 it delivers reasoning at 72, which is a different value proposition entirely if your task needs thinking.
Match the model to the task honestly. If you are processing millions of trivial requests, Nano's pricing is unbeatable and the right call. The moment quality starts mattering, climb one rung to V3 or R1, where a small price increase buys disproportionate capability.
Nano is the actual floor when price is the only axis that matters. V3 and Flash Lite cost more per token but buy back capability you'll miss the moment your prompts stop being trivial.
Last updated June 2026