About AICostLab

Last updated: 2026-05-19

Why we built this

LLM pricing in 2026 is a mess. There are 50+ providers, each with its own pricing model, hidden discounts (prompt caching, batch APIs, context-length tiers), and a constant stream of new model releases. When we tried to estimate "how much would it cost to run my chatbot on Claude vs GPT vs Gemini?" we couldn't find a single tool that gave us a real answer.

Most existing comparison sites just dump a price table and leave you to do the math. They ignore caching savings (which can be 90% off on Anthropic), batch discounts (50% off on OpenAI), and the fact that the same open-source model often costs 3x more on one host than another.

So we built AICostLab to solve our own problem — and made it free for anyone else who needs to make this decision.

What we do differently

• Coverage: 2003+ chat-capable models across 76 providers, refreshed regularly.
• Real cost: we factor in prompt-cache discounts, cache-write fees, batch API rates, and per-token pricing differences between input and output.
• Multi-provider comparison: for open models (Llama, DeepSeek, Qwen, Mistral) we show what each hosting provider charges side-by-side.
• Decision tool, not just a table: tell us your use case and we'll rank the best models for you.
• Updated daily: pricing changes weekly across vendors — stale data leads to bad decisions.

Data sources

Pricing data is sourced primarily from the open-source LiteLLM model prices JSON, a community-maintained file covering 350+ models with multi-weekly updates. We layer on data from the OpenRouter API for multi-provider hosting comparisons on open-weight models.

For new model launches that haven't reached LiteLLM yet, we manually verify against the provider's official pricing page (OpenAI, Anthropic, Google, etc.) and patch the data.

Methodology

Monthly cost estimates assume a workload split (typically 70% input / 30% output for chat use cases) and a cache hit rate (30% default — adjust for your scenario). Numbers are computed at build time from the live pricing data, then statically rendered. There are no servers, no databases, no analytics that watch what you type.

We mark models as deprecated when the provider has announced a retirement date, but we keep the page online so historical decisions can be referenced.

Who we are

AICostLab is built by a solo indie maker with a software engineering background. We're not affiliated with any LLM vendor, hosting provider, or VC. The site runs on Cloudflare's free tier and we keep it that way on purpose — so the data stays honest.

A note on accuracy

We do our best to keep pricing current, but vendors change rates without warning and aggregators occasionally lag. Always verify with the provider's official pricing page before making a business decision. If you spot stale data, tell us — we usually patch within a day.