Anthropic has introduced its new Claude 4 model family. It focuses on two variants: Claude Sonnet 4 and Claude Opus 4. Both models are now available via the Claude website and API. With this release, Anthropic follows recent developments at Microsoft and Google, which recently unveiled their own AI models.
Following OpenAI's GPT 4.1 and Google's Gemini 2.5, Anthropic is now following suit. The Claude models had recently lost some visibility after other providers released new updates. With Claude 4, Anthropic is now offering an upgrade that is likely to attract renewed attention, especially among developers and tech-savvy users. Both Sonnet 4 and Opus 4 demonstrate clear improvements over their predecessors in benchmarks and application scenarios.
Claude Sonnet 4: More power in the mid-range
Claude Sonnet 4 is the mid-range model in the Claude 4 series—between the Claude Haiku (lightweight model) and the Claude Opus (high-end). Compared to the previous version, Claude Sonnet 3.7, Anthropic says there have been some significant improvements. The main improvements include:
- Programming skills
- Logical thinking
- Following instructions
These areas are crucial for developers when working with code, handling complex queries, or solving technical problems. Sonnet 4 is more versatile than its predecessor and offers a good balance between performance and cost.
Claude Opus 4: The new flagship
Opus 4 is Anthropic's most powerful model. In benchmarks for multilingualism, handling of agent-based tools, and reasoning skills, it performs equally well or better than OpenAI's GPT-4.1 or Google's Gemini 2.5 Pro. Opus 4 was able to play Pokémon autonomously for 24 hours—significantly longer than its predecessor, which only lasted 45 minutes. Although Gemini 2.5 Pro recently completed a full Pokémon game, Claude shows a clear improvement in agent-based tasks.
Tools and Memory: New Thinking with Claude 4
An important new feature is the use of tools for extended reasoning. Claude 4 can use tools such as web searches during the reasoning process. It thinks, accesses tools, processes the results, and then continues with the task. This capability is currently in beta testing, but opens up new applications, for example, for research or data-driven decision-making. Both models also demonstrate significantly better memory. If you give Claude access to local files, it can:
- Extract and store information
- Recognize connections
- Building long-term contextual understanding
This means you have to repeat less, Claude remembers relevant content for future interactions.
Claude Code: Developer-friendly and directly integrable
With Claude Code, Anthropic is specifically targeting developers. The feature is now generally available. It supports:
- Background tasks via GitHub Actions
- Native integration in VS Code and JetBrains
- Direct editing of files in the editor
- @mentions in GitHub pull requests
This allows you to integrate Claude directly into your development process – similar to an AI pair programmer who provides input as needed or inserts suggestions directly into your code.
New API features for more control
In addition, the Claude API brings four new features:
- Code execution tool – to execute code directly
- MCP Connector – for connecting to agent systems
- Files API – for direct access to files
- Prompt caching – saves prompts for up to one hour
These new tools make it easier to build complex AI-powered workflows and manage agents more precisely.
Availability and prices
Claude Sonnet 4 and Claude Opus 4 are available through the Anthropic API and through partners such as Amazon Bedrock and Google Cloud Vertex AI. Pricing:
- Claude Opus 4: $15 per million tokens for input, $75 for output
- Claude Sonnet 4: $3 per million tokens for input, $15 for output
Claude 4 in use: Ready to use, versatile
Claude 4 brings noticeable improvements for developers, AI researchers, and tech-savvy users. Claude Opus 4, in particular, demonstrates strong performance in benchmarks, both in natural language processing and agent-based tasks. The new tool-use features, improved memory, and Claude Code significantly expand the application possibilities. If you work with AI—whether for programming, automation, or research—Claude 4 offers you many new options. The models are immediately usable and accessible across multiple platforms. You can get started right away, try out, and test what works best for your workflow. (Image: Shutterstock / Lea Rae)
- Has AI already overtaken humans?
- 184 million passwords leaked – including Apple data
- OpenAI wants to complement smartphones with new AI device
- Apple uses first-person perspective to train robots
- Foxconn: AI and robotics take over the assembly lines
- Google I/O: These are the most exciting announcements of 2025
