Claude Opus 4.5 Arrives: Crushing Coding Tests and Heralding the 'Superman' Era of AI Programming

The pace of large model releases has been relentless lately. Just as Gemini 3 Pro was in the spotlight, Anthropic has officially launched Claude Opus 4.5, once again raising the bar with a strong focus on programming and system-level tasks.

Anthropic claims Opus 4.5 is smarter and more efficient overall. It maintains its top-tier performance in "system-level tasks" like programming, building agents, and computer control, while also showing significant improvements in daily tasks like research, presentations, and spreadsheet analysis. Starting today, Opus 4.5 is widely available through the Claude app, API, and major cloud platforms for developers to use via the `claude-opus-4-5-20251101` API call.

Opus 4.5 Takes Center Stage in a Season of AI Debuts

According to official announcements and tester feedback, Claude Opus 4.5 has a markedly better understanding of ambiguous requests and is more stable in autonomously identifying complex bugs. It has become the first model to score above 80% on the real-world software engineering benchmark, SWE-Bench Verified.

The model's code quality has seen a comprehensive upgrade. In the SWE-bench Multilingual test, which covers eight programming languages, Opus 4.5 achieved the top score in seven of them. In a compelling example, the Anthropic team gave Opus 4.5 a high-difficulty test used for hiring performance engineers. Within the two-hour time limit, the model outscored all human candidates. Beyond software engineering, Claude Opus 4.5 demonstrates across-the-board improvements in vision, reasoning, and mathematics. The model's capabilities are even beginning to outpace existing evaluation standards. In one agentic benchmark, the model devised a clever workaround to a problem that complied with rules but was outside the test's expected answers, showcasing its creative problem-solving abilities.

Claude Everywhere: Integrated into Your Desktop, Browser, and Excel

Alongside Opus 4.5, the entire Claude ecosystem has been upgraded. Claude Code received two major updates: "Plan Mode" for more precise execution plans and a new desktop application for running multiple agentic sessions simultaneously. For app users, the highly requested "endless conversations" feature is now available, allowing dialogues to continue indefinitely by automatically summarizing early context. Furthermore, the Claude for Chrome extension is open to all Max users, and the Claude for Excel beta has been expanded to Max, Team, and Enterprise users.

Smarter and More Economical: A Major Underlying Upgrade for Opus 4.5

As models become smarter, they solve problems more efficiently. Claude Opus 4.5 uses significantly fewer tokens than its predecessors to achieve similar or better results. A new "effort" parameter in the API allows developers to choose between prioritizing speed and cost or maximizing model capability. At a medium effort level, Opus 4.5 matches the best performance of Sonnet 4.5 on SWE-bench Verified but with 76% fewer output tokens. Anthropic also introduced three new features to solve the challenge of managing numerous tools in agentic workflows: "Tool Search Tool," "Programmatic Tool Calling," and "Tool Use Examples," which have significantly boosted accuracy in complex multi-tool tests by reducing token usage and improving tool selection.

This release highlights an emerging trend: different AI models are developing distinct "personalities." The Opus line excels at programming, structured reasoning, and system operations, while models like Sonnet may be more cost-effective for creative writing. The launch of Opus 4.5 confirms this specialization. In the future, selecting an AI model will be less about leaderboard scores and more about finding a "colleague" whose working style aligns with your needs.

Back to blog

Related Posts

  • _{region} ( _{language} )

    • _{area}

    _{region}
    _{language}