Why this is the AI tool I keep coming back to

I’m going to be honest: I was a GPT-4 loyalist for two years. I had the Plus subscription, the custom GPTs, the whole setup. When Claude 4 Sonnet launched, I expected to be underwhelmed — yet another model chasing the leader.

Then I used it for an actual workday. And another. And another. By day five, my GPT tab had been closed and never reopened.

What I use it for

The Claude models have a specific feel: they’re structured in a way that GPT isn’t. When I ask for a brief, I get one. When I ask for a spec, the response looks like a spec. That sounds trivial until you’ve spent twenty minutes asking GPT to “make this into a table please, with columns for X and Y.”

Concrete tasks I’ve handed it in the last week:

• Turning a 3,000-word messy brain-dump into a publishable review outline
• Refactoring a 400-line TypeScript file I’d written six months ago and forgotten
• Generating the schema for our affiliate click-tracking events
• Writing the first draft of three blog posts (I still edit — but the draft is good)

What it’s not great at

Let me be honest about the edges:

• Image generation — Anthropic hasn’t shipped this. If you need image gen, you’ll keep an OpenAI or Midjourney sub.
• Speed at peak — Sonnet gets rate-limited harder than I’d like during US business hours. Not a dealbreaker, but real.
• It will still hallucinate — just less than the competition, in my experience. Always verify code and facts.

The pricing question

$20/mo sounds like a lot. But:

• If it replaces three other AI subs (or three hours of your week), the math is obvious.
• The free tier is genuinely useful — try it before you commit.

I’d start with the free tier for a week. If you find yourself opening it daily, upgrade. You’ll feel the difference in the first paid week.

How I tested it

I used Claude 4 Sonnet exclusively for eight weeks across:

• 3 full product reviews
• 1 production code refactor (the kanban query layer at uzi.network)
• ~40 hours of writing, outlining, and editing
• 200+ code generation tasks

Every claim above is something I actually did. The verdict isn’t from a spec sheet — it’s from the work.

FAQ

Is Claude 4 Opus worth the extra $15/mo? For most people, no. Sonnet is 95% of Opus at 25% of the price. Upgrade only if you regularly hit Sonnet’s context or rate limits.

Does it work offline? No. Cloud-only.

Can I use it commercially? Yes — your inputs/outputs are not used for training by default.