Last updated: 2026-06-12
Mistral AI offers Apache 2.0 open-weight models, a $2/$6 per-million-token API, and the Vibe assistant from $14.99/month with EU-hosted SOC 2 security.
Mistral AI is a Paris-based lab valued at $13.7 billion, offering open-weight models like Mistral Large 3 with 675 billion parameters, Apache 2.0 license, and a 256,000 token context window, alongside a hosted API priced at $2 per million input tokens and $6 per million output tokens. Its Vibe assistant, formerly Le Chat, costs $14.99 per month for Pro.
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. Its flagship Mistral Large 3 is a Mixture-of-Experts model with 675 billion total parameters and 41 billion active parameters per token, trained from scratch on 3,000 Nvidia H200 GPUs and released under the Apache 2.0 license with a 256,000 token context window and a built-in 2.5 billion parameter 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 system with a configurable reasoning dial. 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. API pricing is billed 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, and Ministral 3B costs $0.04 for both. Vibe has a free tier with unlimited rate-limited access to every model, and Vibe Pro costs $14.99 per month, cheaper than ChatGPT Plus at $20. Vibe is available on web, iOS, Android, Windows, macOS and Linux. The May 2026 Vibe update added persistent memory, a built-in code interpreter, and a Work Mode that lets an agent run multi-step workflows across connected tools, drafting messages, creating issues and generating reports with visible tool calls and human approval required for sensitive actions.
Vibe (formerly Le Chat) has a free tier with unlimited, rate-limited access to every model. Vibe Pro is $14.99/month (no included 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.
Mistral AI is a French artificial intelligence lab founded in April 2023 in Paris by Arthur Mensch, Guillaume Lample and Timothee Lacroix, former researchers at DeepMind and Meta. The company builds large language models, both as open weights and as a hosted API, plus a consumer assistant called Vibe, formerly Le Chat. Its flagship model, Mistral Large 3, is a 675 billion parameter Mixture-of-Experts model with 41 billion active parameters and a 256,000 token context window, released under the Apache 2.0 license. Mistral is valued at $13.7 billion as of September 2025. It is widely seen as Europe's leading challenger to OpenAI, Anthropic and Google in foundation models.
Mistral's consumer app, Vibe, has a free tier with unlimited but rate-limited access to every model, and Vibe Pro costs $14.99 per month, which does not include API credits. Enterprise plans are custom-priced and include 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 tokens on Large 3 drop to $0.05 per million, a 90% discount. The hidden cost to watch is that Vibe Pro and API usage are billed independently.
The flagship Mistral Large 3 is an open-weight Mixture-of-Experts model with 675 billion total parameters, 41 billion active per token, a 256,000 token context window, and a built-in vision encoder for images, screenshots and documents. The Vibe consumer app, renamed from Le Chat in May 2026, added persistent memory, a code interpreter, and a Work Mode that lets an agent run multi-step workflows across connected tools with human approval for sensitive actions. Studio and the Agents API support Model Context Protocol connectors for tools like GitHub, Slack and Snowflake. Smaller models like Ministral 3B and Mistral Small 4 are efficient enough to run on a handful of GPUs or locally.
Yes, Vibe has a free tier that gives unlimited access to every Mistral model, subject to rate limits, with no credit card required. The free tier does not include Vibe Pro features like the code interpreter, persistent memory or Work Mode at full capacity, and it does not include any API credits. Developers can also use Mistral's open-weight models, such as Mistral Large 3 under Apache 2.0, completely free by downloading and self-hosting them on their own GPUs, though the largest models require an 8xH200 GPU node. Paid API access starts as low as $0.04 per million tokens on Ministral 3B.
ChatGPT is the best alternative if you want the largest plugin ecosystem and top-tier reasoning benchmarks, 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 is best for developers and enterprises that need GDPR-compliant, EU-hosted LLM infrastructure with SOC 2 Type II and ISO 27001 certifications, and for ML teams that want to self-host an open-weight frontier model under Apache 2.0 rather than depend on a single API vendor. It also suits cost-sensitive startups building agentic apps that use MCP tool calling through Studio Connectors. For example, a European fintech that cannot send data outside the EU could run Mistral Large 3 on its own infrastructure. It is a weaker fit for non-technical users who just want a polished, plug-and-play assistant without configuration.
Yes, Mistral offers a full API through La Plateforme and Studio, covering text generation, vision, code generation via Codestral, and audio via Voxtral TTS. Pricing is per million tokens and ranges from $0.04 on Ministral 3B to $2/$6 on Mistral Large 3. The Agents API and Studio support the Model Context Protocol as a first-class feature, letting developers wire in MCP servers for tools like GitHub, Slack and Snowflake using stdio or Server-Sent Events transports, plus reusable Connectors that package an integration once and reuse it across agents and conversations.
On price, Mistral Large 3 is significantly cheaper, with output tokens at $6 per million versus $10 per million for GPT-5, a 40% saving. On the consumer side, Vibe Pro costs $14.99 per month versus $20 per month for ChatGPT Plus. ChatGPT still leads on raw reasoning benchmarks and has a much larger plugin and integration ecosystem with 900 million weekly users. Mistral's advantage is that its flagship models are open weight under Apache 2.0, so they can be self-hosted, and its infrastructure is EU-based by default, which matters for GDPR-sensitive organizations that ChatGPT's US-hosted infrastructure does not satisfy as cleanly.