Mistral Medium 3.5 Setup Guide: Claude Code, Cursor, Zed, API, Bedrock & Vertex

Mistral just released Medium 3.5, a frontier-class model that's reshaping the economics of AI development. With pricing that undercuts GPT-4 and Claude 3.5 Sonnet, this new model targets long-context reasoning and agentic workflows—exactly what enterprises and AI consulting teams need. Whether you're building AI applications or evaluating models for your AI enterprise stack, here's how to integrate Mistral Medium 3.5 across your favorite development tools.

Why Mistral Medium 3.5 Matters Now

Europe's leading AI lab, Mistral, positioned Medium 3.5 as a direct competitor to OpenAI and Anthropic's flagship offerings. The kicker? Significant cost savings on multi-step reasoning and agentic tasks. If you're running AI consulting projects or deploying AI enterprise solutions, benchmarking this model could unlock real margin improvements. The frontier-model landscape just became more competitive—and that's good news for builders.

Claude Code

Claude Code (via Claude.ai's advanced features) doesn't directly support Mistral models through first-party integration, but you can proxy requests through the Mistral API within your Claude-assisted development workflow.

// Add this to your Claude Code session context
// Use Mistral Medium 3.5 for cost-optimized reasoning tasks

const mistralClient = require("@mistralai/mistralai").default;

const client = new mistralClient({
  apiKey: process.env.MISTRAL_API_KEY,
});

async function reasonWithMistral(prompt) {
  const response = await client.chat.complete({
    model: "mistral-medium-3.5",
    messages: [
      {
        role: "user",
        content: prompt,
      },
    ],
    temperature: 0.7,
    maxTokens: 2048,
  });
  
  return response.choices[0].message.content;
}

Paste this snippet into a Claude Code workspace to start querying Mistral Medium 3.5 for complex reasoning tasks. Perfect for AI development workflows that need cost efficiency without sacrificing quality.

Cursor

Cursor's model selection system now includes third-party endpoints. Configure Mistral Medium 3.5 in your Cursor settings for inline coding assistance powered by a more affordable frontier model.

// .cursor/config.json

{
  "models": [
    {
      "name": "mistral-medium-3.5",
      "provider": "mistral",
      "apiKey": "${MISTRAL_API_KEY}",
      "endpoint": "https://api.mistral.ai/v1",
      "contextWindow": 32000,
      "costPer1kTokens": {
        "input": 0.0007,
        "output": 0.0021
      }
    }
  ],
  "defaultModel": "mistral-medium-3.5",
  "enableLongContext": true
}

After updating your Cursor config, restart the editor. You'll now use Mistral Medium 3.5 for autocomplete and inline refactoring. Great for teams running AI consulting engagements where every token counts.

Zed

Zed's language server protocol supports custom model backends. Add Mistral Medium 3.5 to your Zed configuration for real-time code intelligence.

// ~/.config/zed/settings.json

{
  "language_models": {
    "enabled": true,
    "default_model": "mistral-medium-3.5"
  },
  "features": {
    "inline_completions": true,
    "code_actions": true
  },
  "mistral": {
    "model": "mistral-medium-3.5",
    "api_key": "${MISTRAL_API_KEY}",
    "base_url": "https://api.mistral.ai/v1",
    "temperature": 0.3,
    "top_p": 0.95
  }
}

Zed will automatically detect this configuration on startup. The lower temperature (0.3) makes Medium 3.5 ideal for deterministic code generation—especially useful in AI enterprise environments where consistency matters.

Mistral API (Direct)

The most direct route: call Mistral Medium 3.5 via REST. This is your foundation for any custom AI development workflow.

curl https://api.mistral.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $MISTRAL_API_KEY" \
  -d '{
    "model": "mistral-medium-3.5",
    "messages": [
      {
        "role": "system",
        "content": "You are an AI agent optimized for multi-step reasoning and orchestration."
      },
      {
        "role": "user",
        "content": "Break down this complex workflow into executable steps."
      }
    ],
    "temperature": 0.7,
    "max_tokens": 4096
  }'

This is your baseline for any Python, Node.js, or Go integration. Grab your Mistral API key from their console and start benchmarking Medium 3.5 against your current LLM stack.

AWS Bedrock

Mistral Medium 3.5 is available through AWS Bedrock, giving you enterprise-grade security and compliance for AI consulting and AI enterprise deployments.

import boto3
import json

bedrock = boto3.client('bedrock-runtime', region_name='us-east-1')

response = bedrock.invoke_model(
    modelId='mistral.mistral-medium-3.5-v1',
    contentType='application/json',
    accept='application/json',
    body=json.dumps({
        'messages': [
            {
                'role': 'user',
                'content': 'Orchestrate this multi-step agent workflow.'
            }
        ],
        'max_tokens': 2048,
        'temperature': 0.7
    })
)

result = json.loads(response['body'].read())
print(result['content'][0]['text'])

Use Bedrock when you need VPC isolation, audit logging, or compliance with SOC 2 Type II standards. Ideal for regulated AI enterprise projects.

Google Vertex AI

Deploy Mistral Medium 3.5 on Vertex AI for GCP-integrated ML pipelines and enterprise AI workflows.

from vertexai.language_models import TextGenerationModel

model = TextGenerationModel.from_pretrained('mistral-medium-3.5')

response = model.predict(
    prompt='Design an agentic workflow for data processing.',
    max_output_tokens=2048,
    temperature=0.7,
    top_p=0.95
)

print(response.text)

Vertex AI integration is seamless if you're already running Anthropic or OpenAI models there. Mistral Medium 3.5 fits neatly into existing MLOps pipelines, especially for teams doing AI consulting work across GCP environments.

Benchmarking & Cost Comparison

Mistral Medium 3.5 prices at roughly 70% of Claude 3.5 Sonnet on input tokens and 50% on output—a meaningful difference at scale. For a typical AI enterprise running 10M tokens/month, that's $2,100 saved. Run a two-week benchmark across your core use cases before migrating production traffic.

Next Steps

Start with the direct API or Bedrock integration. Test Medium 3.5 on your longest-context, highest-complexity tasks. The frontier model market is now genuinely competitive, and that leverage belongs in your hands. Whether you're consulting on AI strategy or building production AI applications, understanding Mistral's pricing and performance is table stakes in 2024.