Mistral Large 4 opens as a public API preview, with open weights due in October

Mistral AI launched a public preview of Mistral Large 4 on October 6, 2026. The model is available now through Mistral Studio and the Mistral API. Mistral says the open weights will follow at the end of October; the announcement doesn’t name the license.

What it costs

The announcement lists $1.36 per million input tokens and $4.18 per million output tokens. Mistral’s model docs show those as the original prices, with lower sale prices next to them:

Per 1M tokens	Original price	Sale price shown in the docs
Input	$1.36	$0.68
Cached input	$0.14	$0.07
Output	$4.18	$2.09

The docs give no reason for the discount and no end date. Treat the sale price as temporary until Mistral says otherwise.

Specs

Mistral Large 4
Status	Public preview
Architecture	Mixture of experts: 1.05T total, 52B active parameters
Vision	1.6B-parameter vision encoder; multimodal input
Context window	1M tokens (per the docs)
Open weights	Promised for the end of October 2026

Mistral’s claims

Mistral reports 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4, and says the model solves 93% of the Cybench security challenges. It also claims a narrow lead over OpenAI’s GPT-6 Astra on one visual grounding test, Dense 200 (42% vs 41%). All of these are Mistral’s own numbers.

For scale: on the same DeepSWE v1.1 test, the Grok 4.7 announcement reports 71.0%. Grok 4.7 lists $2 input and $6 output, roughly three times Large 4’s sale price.

What to do

Large 4 is a preview, so the model, its limits and its price can change before general release. If you try it, pin your tests to the current version and check the price page again before you move production traffic.