Introducing Mistral Large 4

What changed

Source summary

Mistral released a public preview of Mistral Large 4, a new flagship multimodal model (1 trillion parameters, 49 billion active) available now via API in Mistral Studio, with full model weights to follow by the end of October 2026. It supports 160+ languages, including all EU official languages, and is priced at $1.36 per million input tokens and $4.18 per million output tokens.

Why it matters to your work

Practical interpretation

Mistral's new flagship model is now accessible via API for developers building multilingual or multimodal AI tools, and its strong EU-language support (160+ languages) may suit organisations working across European markets. It is a developer/API release rather than a consumer app, so most non-technical readers will not interact with it directly yet.

When this could help

Worth evaluating now via the API preview for multilingual (160+ languages, including all EU official languages) or multimodal/document-heavy use cases; worth revisiting once open weights ship if self-hosting or data-sovereignty is a requirement.

Mistral Large 4 is a new open-weight, general-purpose multimodal flagship (Mixture-of-Experts, 1.05 trillion total / 52 billion active parameters, 1.6B vision encoder) with a 1M-token context window, available now in public preview via API, with full weights due by end of October 2026 for self-deployment including private cloud or on-premise.

What to check

• Pricing is $1.36 per million input tokens and $4.18 per million output tokens for the preview API. Source ↗
• Full architecture details and benchmarks are not yet published and will come with the open-weight release at the end of October 2026. Source ↗
• The reduced-moderation version is restricted to selected cybersecurity leaders, vetted partners, and state authorities, not generally available. Source ↗