LLM API price trend by brand
One dot per model: its release date against its currentcheapest blended price per 1M tokens (log scale), colored by the lab that made it. The market's whole story is here — each generation ships cheaper, while flagships hold the top. Click a brand to isolate it; click a dot for its page.
108 models, verified 2026-09-05. Each dot = one base model at its cheapest live offer (blended = input + output per 1M tokens; $0 free tiers excluded). Release dates from the index referential. Prices change hands daily; this page rebuilds with them.
FAQ
Why do LLM API prices keep falling?
Each model generation ships cheaper at equal or better capability: competition between labs, cheaper inference hardware, and distillation of large models into small ones. On this chart the spread runs from $0.049/1M (Mistral Nemo (DeepInfra)) to $60/1M (GPT-6 Astra (OpenAI)) — three orders of magnitude, all currently sold.
Does a model's price change after release?
Rarely, and mostly downward: open-weight models get cheaper as hosts compete, and labs occasionally cut list prices. The dominant effect by far is generational — new models replace old ones at lower prices. This chart plots each model's current cheapest offer against its release date.