About StackProof

StackProof answers a simple question honestly: what does an LLM task actually cost? We publish current API prices for the major models and interactive calculators that turn those prices into a real per-task cost, and show when self-hosting a GPU beats paying per token.

Where the numbers come from

Pricing figures are the standard, non-cached input and output rates published on each provider's official pricing page (OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral). We list the date they were verified and publish the whole dataset as downloadable CC-BY JSON, so you can check every figure. Cached, batch, and long-context tiers differ and are noted, not hidden.

How the calculators work

The cost math is open and deterministic: hosted cost per task is your token profile times the model's published rates; the self-hosting comparison amortizes a GPU's hourly cost over your task volume and checks it against the box's throughput. You put in your own numbers; nothing is fudged. First-party throughput benchmarks on real hardware are on the roadmap and will be published with full methodology when they land, not before.

Who runs it

StackProof is an independent engineering project run under a brand identity rather than a personal one. Credibility comes from open data and re-runnable math, not a byline. Corrections and data requests are welcome at hello@macrothrift.com.

How we're funded

The site is free. Some outbound links are affiliate links and we run display ads. This never changes the numbers; the cost is computed the same regardless of what pays. See our privacy policy for details.