Last updated: 2026-08-19
Mistral AI is a French AI lab whose flagship Mistral Large 3 is a 675-billion-parameter, open-weight Mixture-of-Experts model with a 256,000-token context window. The lab also runs Vibe, a consumer chat assistant formerly called Le Chat, and OCR 4, a multilingual document-intelligence model.
About Mistral AI
Mistral AI is a French artificial intelligence lab founded in April 2023 in Paris by Arthur Mensch, Guillaume Lample and Timothee Lacroix, three researchers who previously worked on large language models at DeepMind and Meta. The company has raised roughly $3.05 billion across multiple rounds, including a $2 billion Series C led by ASML in September 2025 that valued the company at $13.7 billion, making it Europe's best-funded AI startup. Mistral's core strategy is releasing frontier-class models as open weights alongside a hosted API and consumer products, led by the Mixture-of-Experts Mistral Large 3, released under the Apache 2.0 license with a fused vision encoder for reading images, screenshots and documents. Smaller models such as Ministral 3B and Mistral Small 4 run on a handful of GPUs or even locally, and Mistral Small 4 merges the company's former reasoning (Magistral), vision (Pixtral) and coding (Devstral) models into one configurable system. Developers choose Mistral for EU-hosted, GDPR-aligned infrastructure, lower per-token API pricing than US frontier labs, and the option to self-host open-weight models for full data control. Teams building agentic workflows use Mistral's Agents API and Studio Connectors, which support the Model Context Protocol for wiring in tools like GitHub, Slack and Snowflake without custom integration code. Consumers use Mistral's chat app, renamed from Le Chat to Vibe at the end of May 2026, as a lower-cost alternative to ChatGPT, available on web, iOS, Android, Windows, macOS and Linux. The May 2026 Vibe update added persistent memory and a built-in code interpreter, plus Work Mode for running multi-step tasks across connected tools.
Pricing
Vibe Pro costs $14.99 per month and does not include API credits. Enterprise plans are custom-priced. API access is billed per million tokens: Mistral Large 3 at $2 input / $6 output ($0.05 cached input), Mistral Medium 3 at $1 / $3, Mistral Small 3.1 at $0.20 / $0.60, Codestral at $0.30 / $0.90, and Ministral 3B at $0.04 / $0.04.
Key Features
- Mistral Large 3: Open-weight Mixture-of-Experts model with 675 billion total parameters and 41 billion active per token, a 256,000 token context window, and a fused vision encoder, released under Apache 2.0.
- Vibe Work Mode: An agent mode in Vibe (formerly Le Chat) executes multi-step workflows such as drafting messages, creating issues and generating reports across connected tools, with visible tool calls and human approval for sensitive actions.
- MCP Connectors in Studio: Studio and the Agents API support Model Context Protocol connectors for tools like GitHub, Slack and Snowflake, so models call external systems without custom integration code.
- Low-cost per-token API pricing: Mistral's per-token API pricing undercuts the leading closed-source frontier labs on both input and output cost, letting high-volume agentic workloads run at a fraction of the price of proprietary alternatives.
- Self-hostable open weights: Mistral Large 3's FP8 weights run on a single 8xH200 GPU node, letting teams deploy a frontier-class model on their own infrastructure.
- Multilingual vision across the model family: Every current model has vision built into core training rather than bolted on, with strong performance in French, German, Spanish, Arabic, Hindi, Chinese, Japanese and Korean.
- OCR 4: Structure-Aware Document Intelligence: Mistral OCR 4 (June 2026) covers 170 languages with paragraph-level bounding boxes and extracted text; independent annotators preferred it at a 72% average win rate over all tested competitors. Ships as a single container for on-premises sensitive-document deployments; available via Mistral API, Amazon SageMaker, and Microsoft Foundry.
- Vibe: Unified Long-Running Agent: Mistral's Vibe is a single agent for extended multi-step tasks, including inbox and calendar catch-up, deep research, deliverable drafting, and recurring process orchestration, with richer admin controls, scoped API keys, multi-account connectors, and a debugger.
Pros
- Mistral Large 3 ships as open weights, letting teams self-host a frontier-class Mixture-of-Experts model on their own infrastructure instead of paying per-token API fees indefinitely.
- Output token pricing on Mistral Large 3 undercuts both GPT-5 and Claude Sonnet 4.5, and cached-input pricing is priced far below the standard input rate.
- EU-hosted infrastructure is the default rather than an add-on, backed by SOC 2 Type II and ISO 27001/27701 certifications, which simplifies GDPR compliance for European buyers.
- Vibe Pro costs less per month than the leading closed-source chat subscription, and Mistral does not gate its API behind a paid plan: developers can start building on free-tier rate limits immediately.
Cons
- Using the raw API means building your own interface, integrations and analytics; Mistral positions itself as an engine rather than a finished product like ChatGPT or Claude.
- Independent testing found Mistral's coding models handle single-file refactors well but lose consistency across module boundaries on multi-file changes.
- Zero Data Retention is restricted to the Scale plan and stateless API calls only; it does not cover Vibe, agents, conversations, libraries or batch files.
- Vibe Pro's monthly subscription does not include any API credits, so heavy API users pay separately on top of the consumer plan.
Data Handling
- Training-data policy
- Trains on input and output data by default unless Zero Data Retention is activated (Scale plan, stateless API only) or the specific product is opted out of training by default
- Data retention
- 30 days
- Compliance
- SOC 2 Type II · ISO 27001 · ISO 27701 · GDPR
Frequently Asked Questions
What does Mistral AI actually cost?
Vibe Pro costs $14.99 per month, which does not include API credits, and Enterprise plans are custom-priced with on-premise deployment options. API access is billed separately per million tokens: Mistral Large 3 costs $2 for input and $6 for output, Mistral Medium 3 costs $1 and $3, Mistral Small 3.1 costs $0.20 and $0.60, Codestral costs $0.30 and $0.90, and Ministral 3B costs $0.04 for both. Cached input on Large 3 drops to $0.05 per million. The hidden cost to watch is that Vibe Pro and API usage are billed independently.
Can you use Mistral AI without paying?
Yes. Vibe's free tier gives rate-limited access to every model in the lineup and asks for no card. It leaves out Vibe Pro features such as the code interpreter, persistent memory, and full Work Mode capacity, and carries no API credits. Mistral's open-weight models, Large 3 included, also run at no licence cost on your own GPU infrastructure.
What are Mistral AI's closest competitors?
ChatGPT is the best alternative if you want the largest plugin ecosystem and higher reasoning benchmark scores, though it costs more per token and per seat. Claude is the best choice if coding accuracy and agentic task reliability are the top priority, since it scores higher on SWE-bench style benchmarks. Gemini is worth choosing if you are deep in Google Workspace or need its 1 million token context window. DeepSeek is the closest competitor on price and open-source philosophy, undercutting Mistral on per-token cost while Mistral leads on EU hosting and SOC 2 compliance.
Mistral AI or ChatGPT: which should you pick?
On price, Mistral Large 3 undercuts GPT-5 on output tokens by roughly 40%, and Vibe Pro costs less per month than ChatGPT Plus's $20 subscription. ChatGPT still leads on raw reasoning benchmarks and has a far larger plugin ecosystem with roughly 900 million weekly users. Mistral's advantage is that its flagship models ship as open weights that can be self-hosted, and its infrastructure is EU-based by default, which suits GDPR-sensitive organizations better than ChatGPT's US-hosted infrastructure.
How long does it take to get going with Mistral AI?
About five minutes for chat, considerably longer for the self-hosted path. Signing in at mistral.ai or installing the Windows, macOS, Linux, iOS, or Android app needs no card. Developers generate an API key on La Plateforme and follow the quickstart to call Large 3 or Ministral 3B, while teams putting the open weights on their own GPUs should budget hours rather than minutes for that deployment.
Top Alternatives
- ChatGPT: ChatGPT keeps the bigger plugin ecosystem and the top reasoning scores; Mistral AI answers with cheaper tokens and weights you can host yourself.
- Claude: Claude leads on coding and agentic benchmarks, while Mistral AI's case rests on EU data sovereignty and lower output pricing.
- Gemini: Gemini suits teams already inside Google Workspace or needing a 1 million token context window, where Mistral AI offers open-weight deployment flexibility.
- DeepSeek: DeepSeek wins on raw per-token price; Mistral AI holds EU hosting and SOC 2 compliance.
HokAI guides covering Mistral AI
- Best AI Assistant Apps for Android in 2026, Now That Google Assistant Is Dead: Google Assistant shut down on Android September 4, 2026. Here is how Gemini, ChatGPT, Perplexity, Copilot, Meta AI and Claude compare as its replacement.
- Best AI Chatbots in 2026: Pick by the Job, Not the Leaderboard: ChatGPT, Claude, Gemini, Mistral's Vibe, Poe and Character.ai compared on price and actual job fit, not raw model rank, with a decision table and 2026 pricing.