The Claude 4 Family: Three Models, Three Profiles
Anthropic structures its fourth generation around three models with complementary profiles:
- Claude Opus 4.8: the most powerful model, designed for complex reasoning, code analysis, and multi-step agent orchestration. Ideal for architecture reviews or in-depth security analyses.
- Claude Sonnet 4.6: the performance/cost balance for everyday use — code generation, documentation writing, mixed text and image queries.
- Claude Haiku 4.5: the ultra-fast, economical model for high-volume use cases: classification, extraction, short responses in automated pipelines.
Extended Context Window and Long-Term Memory
The Claude 4 family supports contexts of up to 200,000 tokens, roughly equivalent to 500 pages of text. For DevOps teams, this is a game-changer: it's now possible to pass an entire codebase, a complete log file, or multiple Terraform configurations in a single request for joint analysis.
Combined with Opus 4.8's Extended Thinking feature, this context window enables deep causal analysis of complex production incidents without losing track of the reasoning chain.
Tool Use and Autonomous Agents
One of the major advances in Claude 4 is the significant improvement to tool use. Models can now chain parallel tool calls, evaluate their results, and adapt their strategy mid-task.
import anthropic
client = anthropic.Anthropic()
tools = [
{
"name": "run_kubectl",
"description": "Executes a kubectl command on the production cluster",
"input_schema": {
"type": "object",
"properties": {
"command": {"type": "string", "description": "The full kubectl command"},
"namespace": {"type": "string", "description": "The target Kubernetes namespace"}
},
"required": ["command"]
}
},
{
"name": "get_cloudwatch_metrics",
"description": "Retrieves CloudWatch metrics for an AWS resource",
"input_schema": {
"type": "object",
"properties": {
"resource_id": {"type": "string"},
"metric_name": {"type": "string"},
"period_minutes": {"type": "integer"}
},
"required": ["resource_id", "metric_name"]
}
}
]
response = client.messages.create(
model="claude-sonnet-4-6",
max_tokens=4096,
tools=tools,
messages=[{
"role": "user",
"content": "My payment-api service has had a high error rate for 20 minutes. Analyze the Kubernetes pods and CloudWatch metrics to identify the root cause."
}]
)Model Context Protocol (MCP): The Claude Tool Ecosystem
The Model Context Protocol has become the open standard for connecting Claude to external data sources and tools. In 2026, the MCP ecosystem counts hundreds of official and community connectors:
- AWS MCP Servers: native access to CloudFormation, ECS, Lambda, CloudWatch directly from Claude
- Kubernetes MCP: reading cluster resources, analysing events, generating manifests
- Terraform MCP: planning, state diffs, module generation
- GitHub MCP: PR reviews, diff analysis, issue management
- Datadog / Grafana MCP: querying dashboards and alerts in natural language
# Example MCP configuration for Claude Code
{
"mcpServers": {
"aws": {
"command": "uvx",
"args": ["awslabs.aws-mcp-servers"],
"env": {
"AWS_PROFILE": "production",
"AWS_REGION": "eu-west-1"
}
},
"kubernetes": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-kubernetes"],
"env": {
"KUBECONFIG": "/home/user/.kube/config"
}
}
}
}Prompt Caching: Reduce Costs by 90% on Repetitive Contexts
The Prompt Caching feature allows caching large prompt prefixes (documentation, codebase, system instructions) between API calls. For a code analysis pipeline that always passes the same base context, the savings are significant:
| Scenario | Cost without cache | Cost with cache | Savings |
|---|---|---|---|
| 100 code reviews (same 50k-token codebase) | ~$5.00 | ~$0.50 | 90% |
| Chatbot with 10k-token system context | ~$2.00 / 1000 msgs | ~$0.20 / 1000 msgs | 90% |
| Log analysis (same schema repeated) | ~$3.50 | ~$0.35 | 90% |
response = client.messages.create(
model="claude-sonnet-4-6",
max_tokens=1024,
system=[
{
"type": "text",
"text": "You are an SRE expert specialised in AWS and Kubernetes...",
"cache_control": {"type": "ephemeral"} # Cache this system block
}
],
messages=[{"role": "user", "content": user_query}]
)Vision and Image Analysis: Architecture Diagrams and Screenshots
Claude 4 integrates robust multimodal capabilities directly usable by Cloud teams:
- Analyse an AWS architecture diagram exported from draw.io and identify Single Points of Failure
- Read a Grafana dashboard screenshot and describe visible anomalies
- Interpret a VPC network diagram and verify compliance with segmentation best practices
- Extract data from AWS Cost Explorer PDF reports
Concrete DevOps Use Cases with Claude 4
1. Autonomous Incident Response
With Opus 4.8 and a set of MCP tools (kubectl, CloudWatch, PagerDuty), it's possible to build an incident response agent that detects an alert, retrieves logs, identifies the root cause, and proposes a remediation runbook — all in under 5 minutes, without human intervention for the initial diagnosis.
2. Security Code Review (SAST)
Sonnet 4.6 can analyse entire Pull Requests (diff + codebase context) and identify OWASP Top 10 vulnerabilities, exposed secrets, or overly permissive IAM configurations. Integrated into GitHub Actions via the Anthropic SDK, it produces inline comments directly on the PR.
3. IaC Documentation Generation
Haiku 4.5 excels at low-cost generation of documentation for Terraform modules or Helm charts: READMEs, variable descriptions, usage examples. Its low cost makes it ideal for CI/CD pipelines that automatically document on every merge.
4. FinOps and Cost Optimisation
By providing Sonnet 4.6 with a Cost Explorer CSV export and current EC2/RDS configurations, the agent can identify underutilised resources, suggest Savings Plans tailored to the consumption profile, and estimate potential savings with quantified rightsizing recommendations.
Integration with the AWS Ecosystem
Claude 4 is natively available on Amazon Bedrock, allowing AWS teams to use it without managing Anthropic API keys directly — authentication goes through IAM, billing is consolidated on the AWS invoice, and data stays in the chosen region for GDPR compliance.
import boto3
bedrock = boto3.client("bedrock-runtime", region_name="eu-west-1")
response = bedrock.invoke_model(
modelId="anthropic.claude-sonnet-4-6-20251001-v1:0",
contentType="application/json",
accept="application/json",
body=json.dumps({
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 2048,
"messages": [{"role": "user", "content": "Analyse this Terraform plan..."}]
})
)Claude Code: The Agent IDE for Developers
Claude Code is Anthropic's official CLI that exposes Claude 4 directly in the terminal and IDEs (VS Code, JetBrains). For DevOps teams, it offers:
- Reading and modifying configuration files (Terraform, Helm, Kubernetes YAML) with full repo context understanding
- Shell command execution with model feedback for error correction
- MCP integration to directly connect infrastructure tools
- Hooks mode to automate actions at startup, shutdown, or before each command execution
Claude 4 Family Pricing (June 2026)
| Model | Input (MTok) | Output (MTok) | Cache write | Cache read |
|---|---|---|---|---|
| Opus 4.8 | $15 | $75 | $18.75 | $1.50 |
| Sonnet 4.6 | $3 | $15 | $3.75 | $0.30 |
| Haiku 4.5 | $0.80 | $4 | $1 | $0.08 |
For high volumes, Anthropic offers a Batch API at 50% discount for non-real-time processing — ideal for nightly cost analyses or scheduled security audits.
Conclusion
The Claude 4 family represents a qualitative leap for DevOps and Cloud teams. Between autonomous agents capable of managing incidents end-to-end, Prompt Caching that makes large-scale integrations economically viable, and the MCP ecosystem that connects Claude to the entire existing infrastructure, concrete operational use cases are multiplying. Move2Cloud supports its clients in integrating these AI capabilities within their Cloud and DevOps platforms.