Claude Sonnet 4.5 by Anthropic Can Self-Program for Over 30 Hours

In the race for AI dominance, Anthropic has launched Claude Sonnet 4.5, a model the company claims excels in programming. According to Anthropic, Claude Sonnet 4.5 surpasses its predecessor in writing production-ready code, operating computers, performing cybersecurity tasks, and conducting research, modeling, and forecasting in financial services.

"Claude Sonnet 4.5 redefines our expectations — it can engage in autonomous programming for over 30 hours, enabling our engineers to tackle months-long complex architectural projects in significantly less time while maintaining consistency across vast codebases," said Sean Ward, CEO and co-founder of iGent AI, in Anthropic's announcement.

Claude Sonnet 4.5 Excels in Programming

The ability to let AI work independently for such extended periods suggests that Anthropic may be fulfilling the promise of AI as a capable colleague. Claude Sonnet 4.5 outperforms Claude Opus 4.1 — Anthropic’s previously most powerful model released two months ago — across various benchmarks. The company states that, beyond continuous programming, Claude Sonnet 4.5 also excels at tasks in legal, medical, and broader STEM fields.

Claude Sonnet 4.5 has been fine-tuned to avoid flattery, conspiracy theories, power-seeking behaviors, and encouraging delusional thinking. Anthropic describes it as its "most standard-compliant" model.

Priced at $3 per million input tokens and $15 per million output tokens — the same rate as Claude Sonnet 4 — it presents a favorable option for existing users with heavy programming needs, though it remains relatively expensive within the broader generative AI landscape.

Anthropic Announces Additional Features and Tools for Claude

Anthropic has also rolled out product upgrades:

• Saved checkpoints in Claude Code.
• New terminal interface and native VS Code extension in Claude Code.
• New context editing capabilities and memory tools in the Claude API, extending the autonomy of agents.
• Code execution and file creation features in the Claude app.
• Claude for Chrome extension, available to Max-tier subscribers on the waitlist.
• Claude Agent SDK.

Advancements in AI Reflect Gradual Improvements and Increased Specialization

Gartner predicts that generative AI spending will grow by 149.8% year-over-year in 2025, reaching approximately $14.2 billion in end-user expenditure.

As models evolve, differences between them become less pronounced, according to a Gartner research report (subscriber-only link).

"With the release of Claude Sonnet 4.5, Anthropic aims to solidify its leadership in the programming domain," said Arun Chandrasekaran, a Gartner analyst, in an email to TechRepublic. "However, beyond programming, they've also trained the model to perform well in two additional areas — finance and cybersecurity — where they see opportunities. Improvements in these areas are incremental, though."

"Nonetheless, through its success in establishing MCP as an industry protocol for communication between generative AI and agentic AI, Anthropic has added significant capabilities in memory, agent coordination, and autonomy."

Anthropic distinguishes its new model through specialization, particularly in programming and finance — two industries likely to undergo major workforce transformations due to generative AI.

"Claude Sonnet 4.5 indicates Anthropic's strategic shift toward domain-specific specialization in the increasingly competitive generative AI (GenAI) model market," the research report notes.

Notably, the cybersecurity capabilities of Claude Sonnet 4.5 validate the strong market demand for large language models (LLMs) capable of performing threat analysis and security testing, according to Gartner.