Claude 4: Anthropic pushes the boundaries of reasoning with its new model

A leap forward for reasoning

Anthropic has officially launched Claude 4, the latest iteration of its family of artificial intelligence models. This launch marks a significant milestone in the race for language models, with substantial improvements in reasoning, reliability, and agentic capabilities.

The model comes in three versions: Claude 4 Haiku (fast and affordable), Claude 4 Sonnet (balanced), and Claude 4 Opus (maximum performance). The Opus version posts record results on mathematical reasoning and programming benchmarks.

1 million context tokens

One of the most notable advances is the expansion of the context window to 1 million tokens for Opus. This capability allows the model to process entire documents, complete codebases, or lengthy conversations without losing track.

For developers, this means being able to submit an entire project to Claude and obtain consistent analyses across the full source code, without having to split the context into chunks.

“Our goal with Claude 4 was to create a model that does not merely answer, but truly reasons about complex problems” — Dario Amodei, CEO of Anthropic.

Enhanced agentic capabilities

Claude 4 introduces a native “agent” mode, enabling the model to autonomously execute sequences of complex actions. Combined with the Claude Code tool, developers can now delegate complete tasks such as:

• Refactoring entire codebases
• Writing and running automated tests
• Multi-file debugging with autonomous fixes
• Migrating between frameworks and languages

Security and alignment

True to its philosophy, Anthropic has emphasized Claude 4's security. The model incorporates an improved “constitutional AI” system that reduces hallucinations by 40% compared with Claude 3.5 Sonnet, according to the company's internal benchmarks.

The model is also better able to recognize the limits of its knowledge and indicate when an answer requires human verification.

Pricing and availability

Claude 4 is available today through the Anthropic API and on claude.ai. Pricing remains competitive:

• Haiku: $0.25 / $1.25 per million tokens (input/output)
• Sonnet: $3 / $15 per million tokens
• Opus: $15 / $75 per million tokens

What this changes for the market

With Claude 4, Anthropic is positioning itself as a serious competitor to OpenAI and Google DeepMind. The combination of advanced reasoning, the expanded context window, and agentic capabilities makes it a tool particularly well suited to professional use cases and software development.

The coming weeks will tell whether this technical advance translates into widespread adoption, particularly in the enterprise segment, where reliability and security are decisive criteria.