Claude 4 Review: Where It Excels and Where It Falls Short

Claude has been my daily assistant for three weeks: writing articles, refactoring code, analysing data and drafting emails. After hundreds of conversations, here is where it excels, where it frustrates, and who should pay for it.

The strengths

Writing that reads human

Claude remains the best prose writer of the major assistants. It understands voice, rhythm and audience in a way that other models approximate. Give it a rough draft and it returns something you would actually publish — with the caveat that you must review facts yourself.

Genuinely careful reasoning

Asked to work through a messy problem — a broken query, an ambiguous spec, a financial calculation — Claude shows its work before answering. That deliberation is exactly what you want for anything that will be judged later.

Honest about its limits

When it does not know, it tends to say so instead of inventing confidence. That is rare in this category and extremely valuable in professional use.

The weaknesses

Rate limits

Heavy use hits rate limits quickly, especially on the top tier. For long writing sessions this interrupts flow at the worst moments.

Over-conservatism

Sometimes it refuses tasks that are perfectly safe, or hedges answers the way a nervous intern would. A firmer "yes, here is the answer" would be better for many queries.

Price premium

It is not the cheapest flagship. If your workload is huge, the cost difference with competitors is real and worth modelling.

Who it is for

• Writers and editors: best-in-class prose and tone.
• Analysts: clear reasoning you can follow and check.
• Teams with light volume: the limits hurt less when usage is moderate.

The verdict

Claude 4 is the best all-round assistant for thoughtful work I have tested. The writing quality and reasoning justify the price for people who live in text. If your work is more volume-sensitive or you need deep ecosystem tooling, the alternatives are strong — but for pure intellectual work, Claude still leads.