Mistral
FreemiumOpen-weight LLMs from Europe's leading AI lab.
Published 29 May 2026 · Last updated 27 September 2026
Scores
Popularity4/5
Mistral 7B and Mixtral 8x7B were among the most downloaded open-weight models of 2023-2024. The lab is widely recognised as Europe's premier LLM provider, has $14B valuation, and models are available on every major cloud. Strong in enterprise and research circles.
Learning Curve2/5
The la Plateforme API is OpenAI-compatible, so any developer already familiar with the OpenAI Python SDK can switch with a one-line URL change. Running open-weight models locally via Ollama is also beginner-friendly. The main complexity is the breadth of model choices and understanding trade-offs between MoE and dense architectures.
Flexibility5/5
Unmatched spectrum — from 3B edge models to 675B MoE frontier models; open-weight for full fine-tuning and self-hosting; API for managed access; third-party cloud for enterprise compliance. Codestral covers code, Mathstral covers math, Pixtral covers vision. Configurable reasoning effort in Small 4. Few LLM families offer this range.
Performance4/5
Mixtral MoE models punch well above their weight in inference efficiency. Mistral Large 3 is competitive with frontier models on benchmarks. However, on raw capability benchmarks the top Mistral models are generally tier-2 behind GPT-4o and Claude Opus. The efficiency advantage is real and often decisive for cost-sensitive deployments.
Portability5/5
Open-weight models are maximally portable — download once, run anywhere, no ongoing API dependency. Apache 2.0 permits commercial use and modification. Proprietary models (Mistral Large, Pixtral Large) carry some lock-in, but the API is OpenAI-compatible so switching costs are low.
About Mistral
Mistral AI is a Paris-based lab and Europe's leading developer of open-weight language models. Its lineup is led by Mistral Large 3, an open-weight mixture-of-experts flagship of about 675B parameters, with Mistral Medium 3.5 as the multimodal model tuned for agentic and coding work and Mistral Small 4, which folds the former Magistral (reasoning), Pixtral (vision), and Devstral (coding) lines into one hybrid model with a 256k context window. The Ministral 3 family (3B, 8B, and 14B) covers edge and on-device use.
Specialist models round it out: Codestral and Codestral Embed for code completion and code retrieval, OCR models for document parsing, Voxtral for speech transcription and text-to-speech, and Mistral Moderation and Shieldstral for content safety. Weights for the open models are published under Apache 2.0 and run on Ollama, vLLM, llama.cpp, or Hugging Face Transformers.
Developers reach the models through La Plateforme, Mistral's OpenAI-compatible pay-per-token API, which has a free, rate-limited Experiment tier with no card required and also hosts selected third-party models such as Z.ai's GLM. The same models are available on AWS Bedrock, Azure AI, Google Vertex AI, and Together AI, with EU data residency in Paris for teams that need it.
For end users, Mistral's assistant is Vibe (formerly Le Chat, renamed in May 2026), a single agent with Chat, Work, and Code modes and a VS Code extension, offered free with a paid Pro plan.
Key Features
- Mistral Large 3: open-weight MoE flagship of about 675B parameters
- Mistral Medium 3.5: multimodal model tuned for agentic and coding work
- Mistral Small 4: hybrid instruct, reasoning, coding, and vision model with 256k context
- Ministral 3 (3B, 8B, 14B) for edge and on-device use
- Codestral and Codestral Embed for code completion and code retrieval
- OCR, Voxtral speech, and moderation models for specialist tasks
- Vibe assistant with Chat, Work, and Code modes and a VS Code extension
- Self-host via Ollama, vLLM, or llama.cpp, or use Bedrock, Azure AI, Vertex AI, or Together AI
Pros
- Flagship and small models ship as Apache 2.0 open weights, so they can be self-hosted and fine-tuned
- Free Experiment tier on La Plateforme for prototyping without a card
- OpenAI-compatible API means minimal migration effort for teams already using OpenAI's SDK
- Wide cloud coverage: Bedrock, Azure AI, Vertex AI, and Together AI
- Strong European-language performance and EU data residency suit EU-regulated deployments
- Codestral is one of the few models with genuine fill-in-the-middle code completion
Cons
- Mistral Medium is API-only, so the multimodal mid-tier model cannot be self-hosted
- Model lines keep merging and renaming (Small 4 absorbed Devstral and Magistral, Le Chat became Vibe), which makes versions hard to track
- Open Codestral weights use the non-commercial MNPL license, not Apache 2.0
- Smaller fine-tune and tooling ecosystem than Meta's Llama or Qwen
- Self-hosting Large 3 needs multi-GPU infrastructure
Mistral Pricing
Freemium- · Large 3, Small 4, and Ministral 3 weights under Apache 2.0
- · Self-host via Ollama, vLLM, llama.cpp, or Hugging Face
- · Rate-limited access to all API models, including third-party GLM models
- · About 1B tokens a month, no card required
- · Paid plan for the Vibe assistant (formerly Le Chat)
- · Student rate $5.99/month for up to 12 months
- · $0.15/M input, $0.60/M output
- · Hybrid instruct, reasoning, coding, and vision with 256k context
- · $0.30/M input, $0.90/M output
- · Code completion and fill-in-the-middle
- · $1.50/M input, $7.50/M output
- · Multimodal, tuned for agentic and coding work
- · $0.50/M input, $1.50/M output
- · Open-weight MoE flagship
- · 3B: $0.10/M input and output
- · 8B: $0.15/M input and output
- · 14B: $0.20/M input and output
Tech Stacks with Mistral
n8n AI Agent
ProjectBuild an AI agent in n8n's visual editor. The AI Agent node connects a model such as Claude or OpenAI to tools, memory, and a chat window, and the agent decides which tool to call for each request. The tools are n8n's own integrations, your other workflows, and MCP servers, so there is no backend code to write.
Self-Hosted AI with Ollama and Open WebUI
InfrastructureA private, ChatGPT-style assistant on hardware you own. Ollama runs open-weight models such as Google Gemma or Qwen, and Open WebUI gives them a chat interface in the browser with accounts, document search, and model management. Both run in Docker, and nothing leaves your machine.
Local LLM Dev Stack
DeveloperA coding agent that works against a model on your own hardware. A local runtime such as Ollama or LM Studio serves an open-weight coding model such as Qwen or Mistral's Devstral, and an agent such as OpenCode or Cline does the editing. No code leaves the machine and there is no per-token bill.
Tools Related to Mistral
Works well with Mistral(1)
Unsloth supports efficient LoRA/QLoRA fine-tuning of Mistral models on modest hardware.
Alternatives to Mistral(8)
Mistral and Meta Llama are prominent open-weight LLM alternatives. Mistral is Europe's leading open LLM lab with strong EU-language support; Llama has the larger global ecosystem and fine-tune community.
Mistral spans edge models to a MoE flagship with European API and cloud availability; Xiaomi MiMo is an MIT-licensed 1T MoE aimed at agentic tasks. Mistral for EU deployment, MiMo for permissive frontier-scale weights.
MiniMax M3 offers a 1M-token context and native multimodal input; Mistral provides a range from edge models to a MoE flagship with European hosting options. MiniMax for very long context, Mistral for EU availability and size range.
Kimi is Moonshot AI's long-context family with strong agentic coding and Chinese and English performance; Mistral is Europe's open-weight family, available via its API, major clouds, or self-hosting. Mistral suits EU data needs, Kimi long-context coding.
Mistral and DeepSeek are popular open-weight LLM alternatives sharing MoE architecture roots. Mistral is Europe's leading open LLM lab; DeepSeek offers MIT-licensed weights and lower API pricing.
Mistral and Qwen are popular open-weight LLM alternatives with broad model ranges. Mistral leads on European languages and EU compliance; Qwen leads on multilingual Asian tasks and code generation.
