CostPerPrompt

CostPerPrompt tracks live per-token pricing for 232+ AI models and turns those rates into real-world cost estimates through a suite of workload-based calculators. The site is independent and not affiliated with any AI provider, supported by advertising and affiliate links. Pricing data is refreshed automatically from public provider listings — primarily the OpenRouter public model index, cross-checked against vendors' official pricing pages — with a visible last-refresh date on every page.

What It Is

CostPerPrompt is an AI API pricing intelligence tool that aggregates per-token rates across major model providers — including OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, xAI, Alibaba (Qwen), Cohere, MiniMax, Moonshot (Kimi), NVIDIA, and Z.ai (GLM) — and provides calculators that model how those APIs are actually billed in practice. Rather than presenting raw price sheets, the site accounts for the factors that make real costs diverge from naive estimates: growing chat history, prompt caching hit rates, batch processing discounts, and the asymmetry between input and output token pricing.

How the Calculators Work

The site offers eight distinct calculators, each targeting a different workload type:

• API Cost Calculator — compares any workload across all 232 models, with caching and batch discounts applied
• Chatbot Cost Calculator — simulates real conversations with growing history, resent context, and per-provider cache TTLs
• Agent Cost Calculator — models multi-step loops, tool schemas, and retries, which the site notes can make agents cost 10–30× more than naive math suggests
• RAG Cost Calculator — prices indexing, retrieval, and generation separately, including embeddings
• Voice AI Cost Calculator — calculates the full three-part bill: STT + LLM + TTS per minute, per call, and per month
• GPU Rental Pricing — compares H100, A100, and RTX 4090 prices across 10 providers
• Image API Pricing — covers the $0.003–$0.17 per image range with a batch calculator
• Token Counter — paste text and get token counts and costs for any model, running entirely in the browser

How AI API Pricing Works

CostPerPrompt explains the billing mechanics directly on the homepage: providers charge per token (roughly ¾ of a word), with separate rates for input and output. Output is typically 3–5× more expensive than input. Two discounts significantly change the math: prompt caching can cut repeated input costs by up to 90%, and batch processing reduces costs by roughly 50% when latency is acceptable. The site's calculators account for both, which it notes most cost-estimation articles do not — leading to estimates that run 2–3× too high or too low.

Model Coverage and Data Freshness

As of the last recorded refresh (2026-08-02), the site tracks 232 models across 13 vendors. Each model links to a detail page with cost examples. The pricing data is sourced from the OpenRouter public model index and cross-checked against official vendor pricing pages. The site displays the data refresh date prominently and states that pricing discrepancy corrections ship within hours of being reported.

Independence and Transparency

CostPerPrompt is explicitly independent — not affiliated with any AI provider. The site is funded through advertising and affiliate links, which it discloses are always at no extra cost to users and do not affect how prices are ranked or presented. Estimates are described as planning aids rather than quotes, with the provider's official pricing page and the user's own measured token counts treated as the authoritative source.