Token value differs by up to 10x across the Claude lineup. Enter a budget and an input/output split to see exactly how many tokens each model returns per dollar.
Total dollars you want to spend.
Rest of the tokens are billed as cheaper input.
| Model | Input $/1M | Output $/1M | Tokens / $ budget |
|---|
Anthropic prices the Claude API in two streams: input tokens (everything you send — system prompt, conversation history, documents) and output tokens (what the model generates). Output is always pricier, so the real value of a dollar depends on how much of your traffic is generation versus context.
This tool treats your output share slider as the fraction s of every token that is output, and 1 − s as input. The blended price of one token in dollars is:
price_per_token = (s × out_price + (1 − s) × in_price) / 1,000,000
Then the tokens your budget buys is simply budget / price_per_token. Because both per-million rates are constant, the result scales linearly: double the budget, double the tokens. At a 0% output share you see the pure input-token ceiling; at 100% you see the floor, where each token costs the full output rate.
A chat workload that replays long histories can be 90%+ input, so a dollar stretches close to the input-only number. A short-prompt, long-answer workload (drafting, code generation) skews toward output and burns the budget far faster. Slide the control to your real ratio before comparing models — a model that looks expensive on output can win on an input-heavy job. The row with the most tokens for your current split is highlighted automatically.
The figures below are the standard-tier list prices baked into this page: Opus 4.8 and 4.7 at $5 / $25, Sonnet 4.6 at $3 / $15, Haiku 4.5 at $1 / $5, and Fable 5 at $10 / $50 per million input/output tokens. Prompt caching, batch discounts (50% off), and long-context premiums are not applied — this is the headline rate so you can reason about raw token value. For caching and batch math, use the related estimators below.