Mistral AI — research brief

Slug: mistral-ai · Theme: ai-infra · Date: 2026-09-21 (CT)
Sources preference: primary (mistral.ai pricing, about, models, news; docs.mistral.ai) over press.

FACT (primary)

Company

Mistral AI — Paris-founded frontier AI lab (April 2023) building open-weight and commercial models, developer platform (Mistral Studio / La Plateforme APIs), consumer/pro assistant (Le Chat / Vibe), and enterprise private/self-hosted deployments (mistral.ai, about).
Co-founders: Arthur Mensch (CEO), Guillaume Lample (CSO), Timothée Lacroix (CTO) (about).
Mission framing: open, transparent, cost-efficient European alternative; frontier AI “in everyone’s hands” (about).
Key dates on About timeline include: Seed (Jun 2023), Mistral 7B (Sep 2023), Mistral Large (Feb 2024), and later product milestones through 2025 (about).
Homepage / products: Studio (build, deploy, govern agentic systems), self-hosted Studio, Mistral cloud (EU-hosted APIs), and cloud-provider distribution named on homepage: Google Cloud, AWS, Azure, SAP, IBM, Snowflake, NVIDIA, Outscale (mistral.ai).

Series D (company news — FACT claim)

Sep 2026 company news: raised €3 billion Series D at post-money valuation more than €21 billion; called largest equity fundraising round by a European technology company (mistral.ai/news/…).
Led by Samsung Electronics; co-leads Scaleup Europe Fund (EQT) and PSG Equity.
Company claims: operates across 20 countries; supports 125+ global enterprises; named customers include Airbus, ASML, HSBC.
Purpose of round (company): expand frontier research, compute capacity, infrastructure, commercial growth, international footprint; sovereign AI stack (open-weight models + infra + products).

Pricing — Le Chat / Vibe / Studio plans (mistral.ai/pricing — FACT)

Plan	Price (ex-tax)	Notes from pricing page
Free	$0	Limited Vibe (chat, search, creation, coding in terminal); access to Mistral Studio; limited messages/web searches/coding; image gen; $10/mo in API credits; 100+ connectors
Pro	$14.99/mo	Higher limits; all-day coding (CLI/IDE/web); more image gens; $30/mo in API credits; chat + email support
Education / student	$5.99/mo (verified students; normally $14.99)	FAQ
Team	$24.99/user/mo	Collaborative workspace; up to 30GB storage/user; domain verification; data export; higher Flash-answer allowances (e.g. 200/day Team vs 150/day Pro)
Enterprise	Custom	Private deployments, custom models/UI/tools, SLAs, dedicated support — contact sales

FAQ clarifications (FACT — pricing):

Free plan includes limited Vibe + Studio access.
API pricing: most models per million tokens input/output separately. Example on FAQ: Mistral Large $0.5 /M in and $1.5 /M out.
Batch processing: 50% discount; cached input tokens: up to 90% input cost reduction for repeated prompts.
OCR priced per 1,000 pages; speech models per minute; tool APIs per call.
Self-host: open-weight models (e.g. Mistral 7B) Apache 2.0 for research/individual use; commercial deployments require Mistral license.
Studio dashboard tracks API usage (completion, OCR, fine-tuning, etc.).

Model lineup (mistral.ai/models + docs.mistral.ai models overview — FACT)

Generalist: Mistral Large 3 (open-weight multimodal flagship), Mistral Medium 3.5 (agentic/coding multimodal frontier-class), Mistral Small 4 (Apache 2.0 multimodal/multilingual hybrid), Ministral 3 family (3B/8B/14B edge). Specialist: OCR 4.1 (document AI), Voxtral TTS / Mini Transcribe 2 / Realtime / Small (speech), Devstral 2 (agentic coding; docs list deprecation path toward Medium 3.5), Codestral (low-latency code completion FIM), Codestral Embed + Mistral Embed, Mistral Moderation 2, Shieldstral (moderation), Leanstral (Lean 4). Docs also list third-party models available on platform (e.g. Z.ai GLM) — distribution surface, not Mistral-trained IP.

Products / packaging

Le Chat / Vibe: consumer and pro assistant (web/mobile) with connectors, coding (Vibe), research/thinking features — seat/subscription ladder above.
Mistral Studio / La Plateforme: developer production platform — workflows, agents, connectors, experiments/campaigns/judges, hybrid/dedicated/on-prem deploy, observability, guardrails, AI registry (products/la-plateforme).
Enterprise / private: self-hosted Studio, custom models, in-region / sovereign AI framing (homepage + Series D news).

SECONDARY (press — do not treat as audited financials)

CNBC / TechCrunch / EU-Startups coverage of Sep 2026 Series D (~$3.5B / ~$24B USD framing) — corroborates company €3B / >€21B claim but remain SECONDARY for USD conversions and “largest ever” comparisons.
Prior Series C led by ASML (~€1.7B reported in secondary coverage) — SECONDARY unless cited from company primary.

INFERENCE (labeled in canvas where used)

Teaching arc: Free Le Chat/Vibe + Studio land (credits) → Pro/Team seat SaaS + usage-based Studio/API tokens (PLG) → Enterprise private/self-hosted + custom models → hyperscaler & European cloud marketplaces (Azure/GCP/AWS/SAP/IBM/Snowflake/NVIDIA/Outscale).
Cost structure mirrors frontier labs: GPU/training & inference COGS, free-tier demand gen, R&D on open-weight + commercial models, dual GTM (consumer Pro + enterprise sovereign sales), marketplace partner overhead.

Sources

# Mistral AI — research brief **Slug:** mistral-ai · **Theme:** ai-infra · **Date:** 2026-09-21 (CT) **Sources preference:** primary (mistral.ai pricing, about, models, news; docs.mistral.ai) over press. ## FACT (primary) ### Company - **Mistral AI** — Paris-founded frontier AI lab (April 2023) building open-weight and commercial models, developer platform (**Mistral Studio** / La Plateforme APIs), consumer/pro assistant (**Le Chat** / **Vibe**), and enterprise private/self-hosted deployments (mistral.ai, about). - Co-founders: **Arthur Mensch** (CEO), **Guillaume Lample** (CSO), **Timothée Lacroix** (CTO) (about). - Mission framing: open, transparent, cost-efficient European alternative; frontier AI “in everyone’s hands” (about). - Key dates on About timeline include: Seed (Jun 2023), Mistral 7B (Sep 2023), Mistral Large (Feb 2024), and later product milestones through 2025 (about). - Homepage / products: **Studio** (build, deploy, govern agentic systems), self-hosted Studio, Mistral cloud (EU-hosted APIs), and cloud-provider distribution named on homepage: **Google Cloud, AWS, Azure, SAP, IBM, Snowflake, NVIDIA, Outscale** (mistral.ai). ### Series D (company news — FACT claim) - Sep 2026 company news: raised **€3 billion** Series D at post-money valuation **more than €21 billion**; called largest equity fundraising round by a European technology company (mistral.ai/news/…). - Led by **Samsung Electronics**; co-leads **Scaleup Europe Fund** (EQT) and **PSG Equity**. - Company claims: operates across **20 countries**; supports **125+** global enterprises; named customers include **Airbus, ASML, HSBC**. - Purpose of round (company): expand frontier research, compute capacity, infrastructure, commercial growth, international footprint; sovereign AI stack (open-weight models + infra + products). ### Pricing — Le Chat / Vibe / Studio plans (mistral.ai/pricing — FACT) | Plan | Price (ex-tax) | Notes from pricing page | |------|----------------|-------------------------| | **Free** | $0 | Limited Vibe (chat, search, creation, coding in terminal); access to Mistral Studio; limited messages/web searches/coding; image gen; **$10/mo in API credits**; 100+ connectors | | **Pro** | **$14.99**/mo | Higher limits; all-day coding (CLI/IDE/web); more image gens; **$30/mo in API credits**; chat + email support | | **Education / student** | **$5.99**/mo (verified students; normally $14.99) | FAQ | | **Team** | **$24.99**/user/mo | Collaborative workspace; up to **30GB** storage/user; domain verification; data export; higher Flash-answer allowances (e.g. 200/day Team vs 150/day Pro) | | **Enterprise** | Custom | Private deployments, custom models/UI/tools, SLAs, dedicated support — contact sales | FAQ clarifications (FACT — pricing): - Free plan includes limited Vibe + Studio access. - API pricing: most models **per million tokens** input/output separately. Example on FAQ: **Mistral Large** **$0.5 /M in** and **$1.5 /M out**. - **Batch** processing: **50%** discount; **cached** input tokens: up to **90%** input cost reduction for repeated prompts. - OCR priced **per 1,000 pages**; speech models **per minute**; tool APIs **per call**. - Self-host: open-weight models (e.g. Mistral 7B) Apache 2.0 for research/individual use; commercial deployments require Mistral license. - Studio dashboard tracks API usage (completion, OCR, fine-tuning, etc.). ### Model lineup (mistral.ai/models + docs.mistral.ai models overview — FACT) **Generalist:** Mistral Large 3 (open-weight multimodal flagship), Mistral Medium 3.5 (agentic/coding multimodal frontier-class), Mistral Small 4 (Apache 2.0 multimodal/multilingual hybrid), Ministral 3 family (3B/8B/14B edge). **Specialist:** OCR 4.1 (document AI), Voxtral TTS / Mini Transcribe 2 / Realtime / Small (speech), Devstral 2 (agentic coding; docs list deprecation path toward Medium 3.5), Codestral (low-latency code completion FIM), Codestral Embed + Mistral Embed, Mistral Moderation 2, Shieldstral (moderation), Leanstral (Lean 4). Docs also list third-party models available on platform (e.g. Z.ai GLM) — distribution surface, not Mistral-trained IP. ### Products / packaging - **Le Chat / Vibe:** consumer and pro assistant (web/mobile) with connectors, coding (Vibe), research/thinking features — seat/subscription ladder above. - **Mistral Studio / La Plateforme:** developer production platform — workflows, agents, connectors, experiments/campaigns/judges, hybrid/dedicated/on-prem deploy, observability, guardrails, AI registry (products/la-plateforme). - **Enterprise / private:** self-hosted Studio, custom models, in-region / sovereign AI framing (homepage + Series D news). ## SECONDARY (press — do not treat as audited financials) - CNBC / TechCrunch / EU-Startups coverage of Sep 2026 Series D (~$3.5B / ~$24B USD framing) — corroborates company €3B / >€21B claim but remain SECONDARY for USD conversions and “largest ever” comparisons. - Prior Series C led by ASML (~€1.7B reported in secondary coverage) — SECONDARY unless cited from company primary. ## INFERENCE (labeled in canvas where used) - Teaching arc: **Free Le Chat/Vibe + Studio land (credits) → Pro/Team seat SaaS + usage-based Studio/API tokens (PLG) → Enterprise private/self-hosted + custom models → hyperscaler & European cloud marketplaces (Azure/GCP/AWS/SAP/IBM/Snowflake/NVIDIA/Outscale)**. - Cost structure mirrors frontier labs: GPU/training & inference COGS, free-tier demand gen, R&D on open-weight + commercial models, dual GTM (consumer Pro + enterprise sovereign sales), marketplace partner overhead. ## Sources - https://mistral.ai/ - https://mistral.ai/pricing/ - https://mistral.ai/about/ - https://mistral.ai/models/ - https://mistral.ai/products/la-plateforme - https://mistral.ai/news/mistral-makes-sovereign-open-weight-ai-to-frontier/ - https://docs.mistral.ai/getting-started/models/models_overview/ - https://techcrunch.com/2026/09/08/mistral-raises-e3b-as-sovereign-ai-becomes-big-business/ (SECONDARY)