Get started
Record · Provider

Mistral AI

Vitna reads 2 pages from Mistral AI every hour and prices 10 of its products in 2 markets, and 4 more are recorded as withdrawn from sale.

10 products priced · 2 markets · 2 pages read · newest confirmation

Mistral AI on the model tapeEvery providerHow the collector readsmistral.ai ↗

01 / 03Priced now
10 products · 2 markets

What Mistral AI sells that the record prices

Language modelsUSD per 1M tokens9 products
ProductFiguresReadSource
GLM-5.3open weightin $1.40 · out $4.40confirmed 22 Nov 14:08 UTC
Mistral Large 4listin $0.70 · out $2.10confirmed 22 Nov 14:08 UTC
Codestrallistin $0.30 · out $0.90confirmed 22 Nov 14:08 UTC
DeepSeek V4.1 Flash 1017open weightin $0.30 · out $1.20confirmed 22 Nov 14:08 UTC
Ministral 3 (14B)listin $0.20 · out $0.20confirmed 22 Nov 14:08 UTC
GLM-5.3-Flashopen weightin $0.15 · out $0.50confirmed 22 Nov 14:08 UTC
Ministral 3 (8B)listin $0.15 · out $0.15confirmed 22 Nov 14:08 UTC
Mistral Small 4listin $0.15 · out $0.60confirmed 22 Nov 14:08 UTC
Ministral 3 (3B)listin $0.10 · out $0.10confirmed 22 Nov 14:08 UTC
Document extraction1 product
ProductFiguresReadSource
Mistral OCR 4.1Any page, returned as markdown with its tables reconstructedtext from a page $0.004 per pageconfirmed 22 Nov 14:08 UTC

Every figure is a row the boards on /markets carry, at the precision Mistral AI published it, and each product opens onto its whole recorded history with the document behind every stretch. A product belongs to Mistral AI because the register says Mistral AI sells it, never because a board spells the name the same way. Read from the record, so each row carries the read that last confirmed it.

02 / 03Pages read
2 read every hour

The pages of Mistral AI the collector reads

PageReadWhat the register says
docs.mistral.ai/inference/pricingmistral-models-pricingevery hourThe generalist and code model tables of the API pricing page, read separately from the OCR table on the same page. Each row prices input, cached input and output per million tokens, and a row is read only where the model page it links and the name it prints both match the registered row, and only while the page's toolbar is served in dollars, on the Standard tier, with regional inference off, and the page's own data lists the table's rows. Cached input is printed and not registered; batch, priority and the regional surcharge are switches the page applies in the browser, and none is registered.
docs.mistral.ai/inference/pricingmistral-ocr-pricingevery hourThe OCR 4.1 row of the OCR table on the API pricing page, its Input cell, in a section that states its prices per 1,000 pages while the page's toolbar is served in dollars, on the Standard tier, with regional inference off. The figure is divided by 1,000. Document AI, which the page priced as a second meter until 2026-10-05, is no longer printed.

A page is read whole, on its own schedule, and every read is kept with the sha256 of the bytes that came back, so any figure above can be taken back to the document it was read from. A retired page keeps its row and its history and is no longer fetched. Whether each read parsed is on the collector page, which reads it live from the record.

03 / 03Withdrawn
4 products

Products Mistral AI stopped selling

ProductFirst read without it
GLM-5.2
Mistral Document AI
Mistral Large 3
Mistral Medium 3.5

Each of these was on a page a healthy source read, and then a read of that page did not carry it. The date is the first such read, which bounds when sale stopped rather than dating it. The product stays registered, so its page still carries every price it was sold at.