TokenAtlasWhat a token actually costs
Pricing varies by provider, model tier, and whether the tokens are input, output, cached, or reasoning. The token cost calculator handles all of that so you don't have to keep four pricing pages open at once.
Why teams use this
Engineers use it to sanity-check the cost of a new feature before shipping. Founders use it to project AI spend before pricing decisions. Finance teams use it to validate provider invoices.
From calculator to platform
The calculator prices one scenario at a time. Saving scenarios lets you keep a base case, a growth case and a model-swap case side by side, and re-run them when a rate card or your volumes change. Everything stays scenario modelling from numbers you enter — there is no usage ingestion or provider account connection involved.
Cost drivers ranked across providers
Output rate is the single biggest source of variation between models: Claude 3.5 Sonnet bills output at 5x its input rate, GPT-4o at 4x, while DeepSeek V3's output rate is only about 4x its input rate but both figures are far below Sonnet's in absolute terms. Comparing models on the blended 70/30 rate rather than the input rate alone avoids underestimating output-heavy workloads.
Choosing a model from the catalog
Blended at 70/30 across the catalog: DeepSeek V3 $0.519/1M, Llama 3.1 70B $0.653/1M, GPT-4o mini $0.285/1M, Gemini 1.5 Pro $2.375/1M, Mistral Large $3.20/1M, GPT-4.1 $3.80/1M, GPT-4o $4.75/1M, Claude 3.5 Sonnet $6.60/1M, o1 $28.50/1M. Pick the cheapest model that meets the task's quality bar, then confirm with a worked scenario at your real volume.
Reading a token-level estimate correctly
This calculator prices tokens you enter — it does not read a provider invoice or attach to an account. Its output is a modelled per-call and per-month cost, useful for comparing models and sanity-checking a proposed feature, but it will diverge from an eventual bill if the real prompt, completion length or call volume differs from the assumptions entered.
Worked example at two volumes
1,000 input / 300 output tokens on GPT-4o mini: 0.001M x $0.15 + 0.0003M x $0.60 = $0.00015 + $0.00018 = $0.00033/call. At 500,000 calls/month that's $165. The same call shape on Claude 3.5 Sonnet ($3/$15) costs $0.003 + $0.0045 = $0.0075/call, or $3,750/month at the same volume — over 20x the mini-tier cost for identical token counts, purely from rate differences.
Assumptions checklist before comparing two models
— Same input/output token counts entered for both models being compared — Same monthly call volume assumption — Whether either model has a context-length pricing tier that applies at your token count — Whether caching applies to one provider and not the other, since that changes the effective rate
What this calculator does not do
It prices token counts you enter against each provider's published per-1M rate table; it does not connect to your API account, read historical usage, or reconcile against an invoice. For the concepts behind why tokens are billed this way, see the token pricing explainer — this page is the pricing tool, not the tutorial.
Frequently asked questions
- Is the token cost calculator free?
- Yes — completely free, no signup required.
- How current is the pricing data?
- Pricing is updated as soon as providers announce changes. We pull from official price pages, not third-party scrapes.
- Which models are supported?
- Every major model from OpenAI, Anthropic, Google, Mistral, Groq, DeepSeek, Cohere and more. New models are added on launch.