How Does Claude 4 Perform on Real-World Tasks?

TL;DR

Claude 4 performs especially well at extracting facts from long documents, recreating web layouts, and converting analytics data into polished dashboards. Claude Opus 4 still failed a custom chess test involving modified pawn movement, showing that strong coding benchmarks do not guarantee reliable complex logic. Claude Sonnet 4 is free, while advanced Opus access requires a paid plan.

Transcript

Clot 4 is finally here. It's been quite a while since their last update. In fact, Claw 3 came out back in March of 2024. They had Claw 3.7 Sonnet that came out back in February of 25. They just released two new models. Claude Opus 4. Opus is their very best model and Claude Sonnet 4. And I'm going to show you exactly how to use it in this video. It... Read More

Key Insights

• Claude Opus 4 is presented as Anthropic's strongest new model, with claimed strengths in coding, long-running tasks, and agent workflows. Claude Sonnet 4 is described as a significant upgrade over Sonnet 3.7, particularly for coding, reasoning, and precise instruction following.
• Claude 4 models are hybrid models that can provide nearly immediate answers or apply built-in reasoning. Their extended thinking can be combined with tools such as web search, which Claude had added only a month or two before the review.
• Claude Sonnet 4 is available free with usage limits, while Claude Opus 4 and some extended-thinking features require a paid plan starting at $20 per month. A Max plan starts at $100 per month and provides five to 20 times more usage.
• The custom chess coding test failed because Claude Opus 4 could not consistently implement modified movement rules. It rendered the board and pieces and initially moved pawns diagonally, but later pawn actions and movement for several standard pieces stopped working correctly.
• Claude's code revision workflow saves time by modifying relevant portions instead of rewriting the entire program. When the reviewer supplied a screenshot showing the custom chess asset filenames, Claude updated the necessary code sections directly, although the resulting game's logic remained unreliable.
• Claude Opus 4 accurately located Robert's total compensation of 343828.85 inside Nvidia's 180-page 2025 annual report. The PDF had been compressed below Claude's 30-megabyte upload limit, and the requested information appeared on page 53 of the document.
• Claude Opus 4 successfully converted a website banner screenshot into embeddable HTML with inline CSS and JavaScript. It reproduced the formatting effectively and created a graphic substitute for the original website image, producing a result that looked good when downloaded and viewed full screen.
• Claude Opus 4 created a polished visual dashboard from a Google Analytics screenshot. The resulting dashboard contained the correct numbers, used a layout judged better than results from Sonnet 3.7, appeared responsive and mobile-friendly, and could be published for team sharing.

Questions & Answers

Q: What is the difference between Claude Opus 4 and Sonnet 4?

Claude Opus 4 is presented as the most powerful model in the Claude family and is aimed at demanding work such as coding, long-running tasks, and agent workflows. Claude Sonnet 4 is positioned as the successor to Sonnet 3.5 and 3.7, with improvements in coding, reasoning, and instruction following. Sonnet 4 is available free with limits, while Opus 4 requires paid access.

Q: How much does access to Claude Opus 4 cost?

Claude Opus 4 and some extended-thinking capabilities are available through paid access beginning with a $20-per-month plan. The reviewer says this Pro plan can still reach its usage limit during regular Claude use. A Max plan starts at $100 per month and offers five to 20 times more usage for people who use Claude frequently.

Q: How well does Claude Opus 4 handle complex coding tasks?

Claude Opus 4 can generate substantial working code and efficiently revise selected sections, but it did not complete the reviewer's complex logic test reliably. It built a Python chess game with custom assets and initially supported pawns moving diagonally like bishops. During testing, pawn behavior broke and several other pieces stopped moving correctly, causing the reviewer to classify the experiment as a failure.

Q: Can Claude 4 analyze long PDF documents accurately?

Claude Opus 4 performed strongly on the long-document test. The reviewer uploaded a compressed version of Nvidia's 180-page 2025 annual report and asked for a specific director's compensation from page 53. Claude found the requested cash information and returned the exact total compensation of 343828.85. The reviewer still recommends checking extracted information because language models can hallucinate.

Q: What is Claude's file upload limit for large PDFs?

Claude has a stated file upload limit of 30 megabytes in the tested interface. The Nvidia annual report originally measured 50 megabytes, so the reviewer compressed it before uploading. The compressed version retained all 180 pages and the relevant information, allowing Claude Opus 4 to locate the requested director compensation accurately within the document.

Q: Can Claude 4 turn a screenshot into website code?

Claude Opus 4 converted a website banner screenshot into embeddable code containing inline CSS and JavaScript. The generated result could be previewed on canvas or downloaded as an HTML file for full-screen inspection. Although it did not reproduce the original photographic image directly, it created a graphic replacement and preserved the banner's formatting effectively.

Q: Can Claude 4 create dashboards from analytics data?

Claude Opus 4 turned a Google Analytics screenshot from an older website into a simplified visual dashboard designed for team sharing. The reviewer checked the displayed figures against the original source and found every number correct. The dashboard also had a polished layout, appeared responsive and mobile-friendly, and could be published directly rather than only downloaded.

Q: What tools and reasoning features are available in Claude 4?

Claude Opus 4 and Sonnet 4 are described as hybrid models that can answer quickly or use built-in reasoning. Extended thinking can operate with tools such as web search, combining background reasoning with current information retrieval. Claude Code is also generally available for developers, can connect with services such as GitHub, and is offered through the API alongside other capabilities.

Summary & Key Takeaways

• Claude Opus 4 and Claude Sonnet 4 are hybrid models that support quick responses and built-in reasoning. Opus is positioned as the more powerful option, while Sonnet 4 replaces Sonnet 3.5 and 3.7. Extended thinking can work with tools such as web search, and Claude Code is generally available for developers.

• A custom Python chess test exposed Claude Opus 4's limitations with unusual game logic. The model created the interface, loaded the supplied piece images after revisions, and initially allowed modified pawn movement. However, pawn behavior eventually failed, and several other pieces stopped moving correctly, so the test was judged unsuccessful.

• Claude Opus 4 performed much better on document analysis and visual creation. It found the exact director compensation figure inside Nvidia's 180-page annual report, recreated a website banner from a screenshot as downloadable HTML, and transformed Google Analytics data into an accurate, responsive dashboard that could be published and shared.