Is Claude 3.7 Sonnet Good for Writing and Code?

TL;DR

Claude 3.7 Sonnet produces strong, well-structured writing, but the demonstrated coding, interface, logic, and hallucination tests yielded mixed or poor results. Its hybrid design supports quick responses and an extended thinking mode, although extended thinking requires a paid plan. It also lacks web access and deep research, limiting its usefulness for current information and fact verification.

Transcript

Claude finally got a brand new upgrade they just released Claude 3.7 Sonet this has an upgrade from 3.5 Sonet and if you've used Claude before it's one of my go-to large language models especially because of its tone and some functionality that they have this is from a company called anthropic and they don't do releases very often in fact claw 3.5 ... Read More

Key Insights

• Claude 3.7 Sonnet is Anthropic's replacement for Claude 3.5 Sonnet and is presented as a hybrid reasoning model. Users can choose normal operation for quick responses or extended thinking for problems that may benefit from longer reasoning.
• Normal mode is designed for most use cases and delivers answers almost immediately without displaying much thinking. Extended thinking is positioned as the more capable option for mathematics and reasoning, allowing users to see the model's step-by-step thought process.
• Claude 3.7 Sonnet is available to free, pro, team, and enterprise accounts, as well as through the Anthropic API. Extended thinking is excluded from the free tier, so users need at least the professional plan to access it.
• Claude's writing quality is the model's clearest strength in the review. It generated a coherent summary with the requested five bullet points and produced 248 words when asked for 250, demonstrating close adherence to both formatting and length instructions.
• Claude lacks web access and reports an October 2024 knowledge cutoff in the demonstrated real-time query. It also lacks a deep research feature, making it less suitable for requests that require current information or analysis across many websites.
• Claude failed the hallucination test by confidently describing a fabricated lemon cream mango as a real, lesser-known variety. The result shows that polished language does not guarantee factual accuracy, particularly when the model cannot verify claims through web search.
• Claude's extended mode struggled to build a functional chess game. The first web-based result could not load the supplied images or support play, and the Python version initially crashed when a piece moved. Five rounds of revision still left basic rules incorrect.
• Claude's front-end revision improved some mobile centering but created poor desktop formatting. Although Anthropic highlighted stronger front-end development capabilities, the demonstrated one-file solution did not produce a layout that worked well across both desktop and mobile views.

Questions & Answers

Q: What is hybrid reasoning in Claude 3.7 Sonnet?

Hybrid reasoning means Claude 3.7 Sonnet offers two ways to respond. Normal mode is intended for most use cases and provides an answer almost instantly without showing much thinking. Extended thinking spends longer on a problem, displays step-by-step reasoning, and is positioned as the better option for mathematics, reasoning, and coding tasks.

Q: Is Claude 3.7 Sonnet available on the free plan?

Claude 3.7 Sonnet is available across Claude's free, pro, team, and enterprise accounts, and it is also offered through the Anthropic API. However, the extended thinking mode is not included with the free tier. Users who want the longer reasoning option must upgrade to at least the professional plan.

Q: How well does Claude 3.7 Sonnet follow writing instructions?

Claude 3.7 Sonnet followed the demonstrated writing prompt closely. When asked to summarize supplied material in 250 words and include five key points as bullets, it returned exactly five bullet points and a total of 248 words. The reviewer also found its default writing tone clear and less overly promotional than competing models tested previously.

Q: Does Claude 3.7 Sonnet have web access or deep research?

Claude 3.7 Sonnet did not have web access in the demonstrated test and responded that its knowledge cutoff was October 2024. It also lacked a deep research feature that could analyze many websites. These limitations make it a weaker choice for real-time questions, current events, and research that depends on verifying information against online sources.

Q: Does Claude 3.7 Sonnet hallucinate factual information?

Claude 3.7 Sonnet hallucinated during the review's mango-variety test. The prompt listed four mango names, including a fabricated lemon cream mango, and Claude described the invented name as a real, lesser-known variety. The result demonstrates that the model can state unsupported information confidently, so factual claims still require independent verification.

Q: How well did Claude 3.7 Sonnet perform at coding chess?

Claude 3.7 Sonnet performed poorly in the demonstrated chess coding task. Its first web-based version did not load the supplied piece images and was not playable. A Python attempt crashed when a piece was moved. After five rounds of revision, the game still mishandled pawn movement, game-ending conditions, and parts of the interface.

Q: How well did Claude 3.7 Sonnet handle front-end design?

Claude 3.7 Sonnet produced mixed results when asked to repair a responsive website layout from an uploaded image. Its one-file front-end solution centered some elements in the smaller mobile view, but the formatting remained weak and the desktop presentation became significantly worse. Additional prompting might improve it, but the initial revision was not broadly successful.

Q: What is Claude Code and who can access it?

Claude Code is described as Anthropic's first agentic coding tool, created because Claude 3.5 Sonnet had been particularly capable at coding. It is designed to work more natively with code and includes GitHub access inside Claude. At the time covered by the source, Claude Code was available only as a research preview, so access was limited.

Summary & Key Takeaways

• Claude 3.7 Sonnet replaces Claude 3.5 Sonnet and introduces hybrid reasoning. Its normal mode returns answers quickly, while extended thinking is intended for tasks such as mathematics, reasoning, and coding. The standard model is offered across Claude account tiers, but extended thinking is unavailable to users on the free tier.

• The writing test was the strongest demonstration. Given a request for a 250-word summary with five bullet points, Claude produced exactly five bullets and 248 words. The reviewer preferred its default tone to other models and highlighted its ability to follow structural instructions without requiring extensive guidance about style or presentation.

• Practical weaknesses emerged during factual, coding, and front-end tests. Claude invented details about a nonexistent mango variety, needed repeated attempts to create a chess game, implemented basic chess rules incorrectly, and failed to recognize the game ending. A mobile layout revision improved centering but damaged the desktop presentation.