Anthropic’s Latest Claude 4 AI Models Excel at Multi-Step Reasoning

23.05.2025

During its inaugural developer conference, Anthropic introduced two new AI models, Claude Opus 4 and Claude Sonnet 4, which it claims rank among the industry’s top performers on popular benchmarks. Both models are designed to handle complex, long-horizon tasks and analyze large datasets, with particular strength in programming tasks, making them ideal for writing and editing code. Sonnet 4 is accessible to both free and paying users, while Opus 4 is exclusive to paying customers, with pricing varying depending on usage.

Anthropic aims to significantly grow its revenue by 2027, targeting $12 billion, up from an estimated $2.2 billion this year. The company has secured substantial funding from investors like Amazon to support the rising costs of developing advanced AI models. Despite strong competition from companies like OpenAI and Google, Anthropic is advancing its AI capabilities with Claude 4, emphasizing its commitment to staying competitive.

Opus 4 stands out for maintaining “focused effort” across multiple workflow steps, while Sonnet 4 serves as an improved replacement for previous models, particularly excelling in coding and math tasks. Both models reduce issues like “reward hacking,” where AI shortcuts tasks to meet objectives. While Opus 4 outperforms some rivals on coding benchmarks, it does not yet surpass others in broader multimodal or advanced scientific tests.

Anthropic has implemented enhanced safeguards for Opus 4 due to its increased potential for misuse, particularly in sensitive STEM-related areas. The Claude 4 models feature a hybrid design that balances rapid responses with deeper reasoning capabilities. Users receive summarized insights into the models’ thought processes, which remain partially hidden to protect Anthropic’s competitive edge. Additionally, the models can multitask, using external tools and storing knowledge to improve reliability over time.

To better support developers, Anthropic upgraded Claude Code, an AI coding assistant that integrates with popular development environments like VS Code and GitHub. Despite ongoing challenges in AI-generated code quality and security, Anthropic commits to more frequent updates to enhance model performance and deliver new capabilities faster, ensuring users stay at the forefront of AI innovation.