Show HN：CostPerPrompt – AI API 实时定价与实际工作负载成本计算器

Show HN：CostPerPrompt – AI API 实时定价与实际工作负载成本计算器
Show HN: CostPerPrompt – Live AI API pricing and real-workload cost calculators

该平台提供了一套全面的计算器，旨在精准估算人工智能的工作负载成本。平台实时更新超过 232 种模型的定价，为您解析聊天机器人、多步智能体、RAG（检索增强生成）流水线、语音 AI 以及图像生成等复杂应用中的各项费用。与通用的估算方式不同，这些工具考虑了提示词缓存（最高可节省 90% 费用）和批量处理（五折优惠）等关键变量，这些因素常被忽略，却是实现精准预算必不可少的环节。此外，该网站还提供 GPU 租赁对比、基于浏览器的 Token 计算器以及实用的成本削减指南。通过模拟真实的业务场景（如对话历史增长或多步智能体循环），该平台能为您提供一份切实可行的人工智能实施“三部分账单”。它是您解读 AI 行业复杂定价模式的必备资源，确保您准确掌握输入/输出比例、缓存及批处理如何影响您的每月最终开支。

**CostPerPrompt** 是一款提供实时 AI API 定价和真实工作负载成本计算工具的新平台。其创始人 Ahmed Hassan 设计该平台的初衷，是帮助用户估算实际支出，并将静态价格表常忽略的变量（如缓存命中率）纳入考量。该项目在 Hacker News 上引发了褒贬不一的评价。一些用户质疑“AI 生成”网页界面的实用性，并指责作者使用 AI 撰写回复；而另一些用户则围绕 AI 成本建模的复杂性进行了技术探讨。关键的技术反馈集中在因供应商行为不一致（例如提示词缓存的生存时间 TTL 设置各异）而导致的成本计算难题上。例如，20 分钟的间歇可能在 OpenAI 中产生缓存命中，但在 Anthropic 的默认设置下却会导致缓存缺失。对此，作者计划增加一项“消息平均间隔”功能，以便用户根据会话行为更好地模拟现实中的成本波动。

原文

Live pricing for 232+ models, refreshed automatically — plus calculators that turn token prices into real answers: what your chatbot, agent, or API workload will cost per month.

Every cost question, answered

$_ API Cost Calculator Any workload, compared across all 232 models — with caching & batch discounts. 💬 Chatbot Cost Simulates real conversations — growing history, resent context, cache hits. 🤖 Agent Cost Multi-step loops, tool schemas, retries — see why agents cost 10–30× more than you think. 📚 RAG Cost Indexing, retrieval and generation priced separately — spoiler: embeddings are pennies. 🎙️ Voice AI Cost STT + LLM + TTS per minute, per call, per month — the full three-part bill. 🖥️ GPU Rental Pricing H100, A100, RTX 4090 across 10 providers — the same chip at a 5× price spread. 🖼️ Image API Pricing $0.003 to $0.17 per image — three market tiers and a batch calculator. 🔢 Token Counter Paste text, get tokens and cost on any model — 100% in your browser. ⚡ Cost-Cutting Guides Prompt caching (−90% input) and the Batch API (−50% everything) explained.

Flagship model pricing

per 1M tokens · updated 2026-08-02

See the full table of 232 models →

How AI API pricing works — the 60-second version

Every major AI provider bills the same way: you pay per token (roughly ¾ of a word), with separate rates for input (what you send) and output (what the model writes back). Output is usually 3–5× more expensive than input. A model listed at $5 / $25 per million tokens costs $5 for every million tokens you send and $25 for every million it generates.

Two discounts change the math dramatically: prompt caching cuts repeated input costs by up to 90% (critical for chatbots that resend conversation history), and batch processing takes ~50% off when you can wait for results. Our calculators account for both — most "how much will this cost" articles don't, which is why their estimates run 2–3× too high or too low.