THE TICKER · MISTRAL AI · CHECKED DAILY · LAST CHANGE 2026-11-10

MISTRAL API PRICING

General availability turned the launch discount into the list price.

Mistral Large 4 waited a month to reach this board, on purpose. It went on sale on October 6, 2026 as a public preview, and while it was in preview Mistral's own model page listed $1.36 in / $4.18 out per million tokens. Price datasets meanwhile carried a launch rate of about half that, so the list price and the price people paid did not match. We hold preview models back for exactly this reason: a preview price is the one most likely to move. On November 5 the model went generally available, and the page now lists $0.70 in / $2.10 out. The cheap launch rate became the list price, rather than the list price coming back after the launch. At 3x, its output multiple is lower than Claude's 5x on every model. Cached reads cost $0.07, a flat 10% of input, and Mistral's caching docs say tokens cache in 64-token blocks. The model page lists batch processing but gives no batch rate, so we show it as not published. The weights are open under Mistral's own license, so the same model can also be run elsewhere, at whatever that host charges. Mistral is also one of the ten providers in our Cost Transparency Index, where it scored 12 out of 25 and ten criteria could not be checked from public pages, mostly around invoices. It did not answer our right of reply.