Tokens per dollar calculator

Most pricing tools answer "what will this cost?". This one answers the question you ask when the budget is the fixed thing: how much work does my money actually buy?

0
total tokens for your budget

Token yield leaderboard

How many tokens one dollar buys on each model, at the default 3:1 input-to-output mix. The twelve highest-yield models tracked here.

# Model Provider Tokens per $1 Blended / 1M
1 Qwen3.7 Flash Alibaba 18,181,818 $0.055
2 Llama 3.1 8B Instant (via Groq) Groq 17,391,304 $0.0575
3 Nova Micro Amazon 16,326,531 $0.0613
4 Command R7B Cohere 15,238,095 $0.0656
5 Nova Lite 1.0 Amazon 9,523,810 $0.105
6 Llama 4 Scout Meta 7,407,407 $0.135
7 Step 3.5 Flash StepFun 6,666,667 $0.15
8 GPT-4.1 nano OpenAI 5,714,286 $0.175
9 Gemini 2.5 Flash-Lite Google 5,714,286 $0.175
10 DeepSeek V4 Flash DeepSeek 5,714,286 $0.175
11 Sonar Small Online Perplexity 5,000,000 $0.20
12 Mistral Small 4 Mistral 3,809,524 $0.2625

How this is calculated

Your ratio sets how the budget splits between input and output. At a ratio of 3, every output token is accompanied by three input tokens, so the effective cost per token is (3 × input rate + 1 × output rate) ÷ 4. Dividing your budget by that gives the total token count, which is then split back out into its input and output halves.

Set the ratio to a large number to model a prompt-heavy workload such as document analysis or retrieval, or to a small one for generation-heavy work such as long-form writing. The gap between those two cases on the same model and the same budget is usually startling, and it is the main reason single-number model rankings mislead.

Why more tokens is not the same as more value

A high token yield only matters if the model uses its tokens well. A cheap model that needs three attempts, or that produces four hundred words where a better one produces a hundred, can burn through its yield advantage entirely, and you pay for the extra output at the output rate, which is the expensive one. Use this page to size a budget and set expectations; use real evaluation on your own task to choose the model.

Tokens per dollar questions

How many tokens can I get for $1?

It ranges from roughly 14,815 tokens on the most expensive model tracked here to about 18,181,818 on the cheapest, at a 3:1 input-to-output mix. That is the same dollar buying wildly different amounts of work depending on which model you point it at.

How many tokens is a page of text?

Roughly 500-750 tokens for a page of English prose, or about 4 characters per token as a working rule. Code, non-English text, and heavily formatted content all tokenise less efficiently. The token counter gives a closer estimate from a real sample.

Why does the input-to-output ratio change the answer so much?

Because output typically costs four to six times what input does. A budget spent entirely on input buys several times more tokens than the same budget spent on output, so a model that looks cheap on a prompt-heavy workload can look expensive on a generation-heavy one.

Is tokens-per-dollar a good way to choose a model?

It is a good way to size a budget and a poor way to choose a model on its own, because it says nothing about how many tokens each model needs to do the job. A model that answers in 200 tokens at twice the rate beats one that rambles for 800 tokens at half the rate.

Related calculators and pages

Rates last verified 2026-08-16. Token counts vary by tokeniser, so the same text yields different counts on different model families. See the methodology page.