What happened on 22 May 2025
Anthropic launched Claude 4, the new generation of its AI model. Two versions are available: Opus 4, the most powerful model, and Sonnet 4, faster and more economical. The major new feature: these models can work autonomously for several hours on complex tasks — writing code, analyzing documents, preparing reports — without human intervention.
What this changes for you
✦ The opportunity
Imagine entrusting an AI with writing a complete 30-page analysis report from your internal data. Or asking it to audit your source code overnight and deliver a vulnerability report to you in the morning. That is exactly what Claude 4 enables in autonomous mode.
The most immediate use cases for an SMB or a mid-market company:
Software development
Claude Code makes it possible to code complete features autonomously. Your developers supervise instead of writing everything.
Document analysis
Processing contracts, tenders, summarizing financial reports — tasks that used to take days are done in hours.
Advanced customer support
AI agents capable of resolving complex tickets end to end, not just redirecting to a human.
⚠ The risk
A model that works alone can also go wrong alone
A model that works alone for hours can also go wrong for hours. Anthropic has built in safeguards (the model stops and asks for confirmation when in doubt), but the risk of "prolonged hallucination" exists. Without appropriate supervision, you could end up with a voluminous deliverable… that is wrong.
Vendor lock-in risk
Building your critical processes on a single proprietary model exposes you to changes in pricing, usage policy or availability.
Our recommendation
Test Claude 4 on a concrete but non-critical use case — for example, generating technical documentation or analyzing your HR data. Measure the real time savings before integrating it into a critical business process.
Identify a repetitive task
A task that consumes more than 4h/week in your team — that is where you will find the best return on investment.
Test for 2 weeks
With systematic human validation of the results — don't let the AI operate without safeguards from the outset.
Measure the ROI
Time saved vs subscription cost vs error rate. The numbers must speak for themselves before scaling.
In summary
Frequently asked questions
What is Anthropic's Claude 4?
Claude 4 is the fourth generation of the artificial intelligence model developed by Anthropic. It comes in two versions: Opus 4 (the most powerful, optimized for long and complex tasks) and Sonnet 4 (faster and more economical, suited to everyday tasks). The major new feature is its ability to work autonomously for several hours without human intervention.
What is the difference between Claude Opus 4 and Sonnet 4?
Opus 4 is Anthropic's frontier model: it excels at complex reasoning, coding and long tasks. Its cost is $15/$75 per million tokens. Sonnet 4 is faster and 5 times cheaper ($3/$15 per million tokens), ideal for writing, summarization and classification. Both share a 200K-token context window.
How can Claude 4 be used in business?
Start by identifying a repetitive task that consumes more than 4h/week (document analysis, report generation, software development, customer support). Test Claude 4 on this use case for 2 weeks with systematic human validation. Measure the ROI (time saved vs cost vs error rate) before integrating the tool into a critical business process.
How much does Claude 4 cost?
Claude 4 is accessible via the Anthropic API, AWS Bedrock and Google Vertex AI. Opus 4 costs $15 input and $75 output per million tokens. Sonnet 4 costs $3 input and $15 output per million tokens. A Claude Pro subscription (conversational use) is also available at $20/month.
For tech profiles
The models in detail
Frontier model
Extended context, optimized for long tasks and complex reasoning. Ideal for code, analysis and planning.
Value for money
Faster, cheaper. Excellent for everyday tasks: writing, summarization, classification.
Claude Code — the key tool for dev teams
- Navigate a full codebase and understand the architecture
- Write, modify and refactor code across multiple files
- Run tests and fix errors in a loop
- Create Git commits with relevant messages
- Connect to external tools via the MCP protocol (databases, APIs, cloud services)
API & Pricing
Quick comparison
| Criterion | Claude 4 Opus | GPT-5 | Gemini 3 |
|---|---|---|---|
| Long-running autonomy | Excellent | Good | Good |
| Code | Excellent | Very good | Good |
| Cost | High | High | Medium |
| Open source | No | Partial (gpt-oss) | No |