Claude Sonnet 4.5: Anthropic’s New Coding Powerhouse with 30-Hour Autonomous Operation
Anthropic's Claude Sonnet 4.5 leads SWE-Bench with 69.8% score, offers extended autonomous operation, and costs $3/$15 per million tokens.
Anthropic has released Claude Sonnet 4.5, positioning it as the world’s best coding model with significant improvements in autonomous operation, security engineering, and enterprise capabilities. The model, launched on September 29, 2025, represents a major advancement in AI-assisted software development and long-running task execution.
On the SWE-Bench coding benchmark, Claude Sonnet 4.5 achieves a leading score of 69.8%, narrowly edging out competitors like GPT-5 Codex (69.4%) and GPT-5 (68.8%). This performance makes it the top-ranked model for real-world GitHub issue code patching tasks, demonstrating superior reasoning and agentic tool use capabilities.
One of the most impressive features is the model’s extended autonomous operation, which supports more than 30 hours of continuous work on complex tasks. The model maintains clarity, focus, and incremental progress while providing fact-based progress updates throughout long-running workflows. This capability enables sustained execution of software projects, financial modeling, and cybersecurity tasks without human intervention.
Pricing for Claude Sonnet 4.5 remains competitive at $3 per million input tokens and $15 per million output tokens. The model offers significant cost-saving features including up to 90% savings with prompt caching and approximately 50% savings with batch processing. While more expensive than GPT-5 models ($1.25/$10 per million tokens), it’s substantially cheaper than Claude Opus ($15/$75 per million tokens) and GPT-4.5 ($75/$150 per million tokens).
Enterprise security features have been significantly enhanced, with Claude Sonnet 4.5 demonstrating improved vulnerability detection and more robust security engineering practices. According to Anthropic’s internal testing, the model reduces average vulnerability intake time by 44% and increases detection accuracy by 25%. The model can deploy autonomous agents that patch software vulnerabilities before exploitation, shifting from reactive to proactive security postures.
The model is available through major enterprise platforms including Google Cloud Vertex AI, Amazon Bedrock, and Snowflake Cortex AI, providing secure API integrations for enterprise workflows. Additional features include a 200K token context window, parallel tool usage, enhanced context management, and improved resistance to prompt injection attacks.
With its combination of top-tier coding performance, extended autonomous operation, competitive pricing, and enterprise-grade security, Claude Sonnet 4.5 establishes itself as a formidable competitor in the AI coding assistant market, particularly suited for complex, mission-critical development workflows.
