Claude 3.5 Sonnet: Performance Review and Real‑World Tests

Claude 3.5 Sonnet, the newest model from Anthropic, is marketed as faster, cheaper, and the strongest globally, with many benchmarks indicating it outperforms GPT‑4o on key metrics.

Independent users tested the model on a common task—generating UI code from a single‑sentence prompt. While GPT‑4o returned code without detailed explanations, Claude 3.5 Sonnet produced complete, well‑matched UI code with additional design details.

The model’s knowledge cutoff was updated to April 2024, allowing it to answer recent events such as the February Super Bowl result.

In Chinese‑language evaluations, Claude 3.5 Sonnet completed a ten‑line story‑writing task ending each line with the word “apple,” and solved a challenging Alibaba math‑competition question without provided options.

Visual reasoning capabilities were highlighted, with users generating chip‑design flowcharts and creating games from a single screenshot in as little as 25 seconds, including a full‑featured Mancala web app.

Claude 3.5 Sonnet also demonstrated strong coding abilities, passing 64 % of internal pull‑request test cases (versus 38 % for Claude 3 Opus) and fixing code errors within seconds.

Users discovered new O(n) sorting algorithms and used the model’s Artifacts feature to run and iterate code interactively, noting a ten‑fold efficiency boost over GPT‑4o and other LLMs.

Despite impressive performance, the model still fails on simple tasks such as playing tic‑tac‑toe or solving basic math word problems, with similar failures observed in Gemini 1.5 pro.

Anthropic’s background is described: founded by former OpenAI veterans, it received heavy investment from Amazon and released Claude 3 in March, which surpassed GPT‑4 across benchmarks. Claude 3.5 Sonnet is the first large‑cup model in the series, with larger variants (Haiku, Opus) planned.

The article concludes with community excitement about Claude 3.5 Sonnet’s dominance and speculation about future releases.