Drawpie
Game guides & tools
AI models · 2026 tool

LLM API Cost Calculator

Put in your monthly token usage and see what Kimi K3, GPT-5.6 Sol and Claude Fable 5 would each cost — on their official list prices. It shows cost only: there's no honest "cost per successful task" to compute, because no benchmark measures all three the same way. The capability index is a reference, never mixed into the price.

01

Your monthly usage

=

What you'd pay

Per model

i

How these three price up

The same 3.3× ladder

Per 1M tokens: Kimi K3 $3 in / $15 out, GPT-5.6 Sol $5 / $30, Claude Fable 5 $10 / $50. Fable costs 3.3× Kimi at every volume — the ordering never changes.

Output dominates

Output tokens cost 5× input for all three, so a chatty, long-answer workload pays far more than a read-heavy one. Your input:output mix matters more than the headline input rate.

Caching is the wildcard

A cache hit bills input at $0.30 / $0.50 / $1.00. Kimi's 90% discount is the steepest, so if your prompts repeat, its lead widens — model your real cache-hit rate above.

Full breakdown with charts: Kimi K3 vs Fable 5 vs GPT-5.6 Sol: price and benchmarks. Also: why cross-lab benchmark comparisons mislead.

This calculator does arithmetic on the numbers you enter, using each lab's own list prices (per 1M tokens), read 18 July 2026: Kimi K3 $3 input (cache-miss) / $15 output / $0.30 cache-hit; GPT-5.6 Sol $5 / $30 / $0.50; Claude Fable 5 $10 / $50 / $1. Caching applies to input only. The "Index" column is Artificial Analysis's composite Intelligence Index (an aggregator score of many evals, all three within ~3 points) — it is a capability reference, not a task-success rate, and is never folded into cost. We do NOT compute a "cost per successful task" or any value-adjusted figure, because no per-model success rate exists across all three on one methodology. Batch/lower tiers exist and aren't modelled. LLM prices change frequently.
Independent tool. Not affiliated with, endorsed by, or associated with Moonshot AI, OpenAI or Anthropic. Model and product names are trademarks of their respective owners.