Hey Product Hunt! 👋 I got tired of looking at flat "cost per million tokens" tables when trying to choose between different LLMs. They just don't reflect how we actually build apps anymore. So I built a simple calculator to model real-world costs: Cascades: What you actually save by routing simple queries to GPT-4o mini and escalating to Sonnet/o3 on failure (including the double-billing overhead of retries). Agent loops: How cost compounds over multi-turn agentic steps, and how much prompt caching actually saves you. Effective Cost: Scaling prices by benchmark success rates. If a cheap model fails half the time, it actually costs twice as much in production. It has the latest models loaded (DeepSeek R1, Grok 3, Llama 4, Gemini 2.5, etc.) and a Cost vs. Intel scatter plot. It's 100% free. Hope this is useful to anyone else trying to make sense of their API options!

TrueLLM Cost Simulator

Simulate LLM costs for cascades, caching, & agent loops

Simulate LLM costs for cascades, caching, & agent loops