New on 3 November 2026: the token market table
Contract full-stack and DevOps engineer
I’m Lance Keay. I build web applications and run the servers and pipelines under them. Here I also review AI tools and self-hosted software I have run myself, verdict first.
Latest verdictReview
Ollama for one person; vLLM once a team shares the GPU.
Latest posts, verdict first
-
Ollama, llama.cpp and vLLM on one GPU
Verdict: Ollama for one person; vLLM once a team shares the card.
-
Restic backups to a second box
Verdict: worth it. One binary, encrypted repositories, restores that work.
-
Open WebUI for a small team
Verdict: good on your own hardware; read the licence before you rebrand it.
AI token market, November 2026
Takeaway: compare cached-input and batch rates, not only list prices.
Monthly price table for AI model APIs
On the first Monday of each month I put the list prices of the main model APIs into one wide table.
- Prices per million tokens4 rates
Input, output, cached input and batch, in US dollars
- Providers5
Anthropic, OpenAI, Google, Mistral and DeepSeek
- Changes since last monthMonthly
Marked in the table, with the old price beside the new
