Top pick, per month
Per message
with the top pick
Messages per month
Tokens sent per message
average, before caching
Our recommendation
Based on your knowledge base size and usage. Prices update as you move the sliders.
Monthly cost by model
Hover or tap a bar for the full breakdown. Striped bars can't fit your full knowledge base, so they're priced with retrieval instead.
Where the money goes
Every model, side by side
Click a column to sort. List prices are per 1 million tokens.
How the math works
- One token is about three quarters of a word, so we use 1.33 tokens per word and 500 words per page.
- Every message sends your instructions, the knowledge text, the chat so far, and the new question. The answer and any hidden thinking bill as output.
- With caching on, the parts that repeat (instructions, full knowledge base, earlier messages) bill at each provider's cached rate, and we assume 10% of those reads miss the cache. Anthropic cache misses bill at 1.25x to cover the cache write.
- We add 300 tokens per message for platform overhead like formatting and tool definitions.
- Your third-party platform may add its own markup or seat fees on top. These are raw model costs.
Where the prices come from
- List prices checked October 1, 2026 from Anthropic, OpenAI, Google Gemini, xAI, DeepSeek, Moonshot, Z.AI, Mistral, Alibaba Cloud, and Groq.
- Open-weight models can also run on your own servers or another host, so their price depends on where you run them. Rates shown are the maker's own API, or Groq where noted.
- Gemini 3.8 Flash uses Google's introductory rate. Grok 4.1 Fast uses xAI's last published rate.
- Tiers are our own grouping: Frontier is each lab's top model, Strong handles most business work well, and Value is built for speed and price.
- AI prices drop fast. Check the provider's page before you sign anything.
Built by The Uncommon Business.