Ask five people which AI assistant they rely on and you’ll get five confident answers, each delivered as though the question were settled. It isn’t. ChatGPT, Claude, Gemini, Perplexity and a growing pack of open-weight models share a lot of DNA and diverge sharply on everything that matters day to day.
So this isn’t another explainer on what ChatGPT is. It’s a comparison of the realistic alternatives, with the trade-offs left in rather than smoothed over. The choice usually comes down to four things: how you write, how you research, how much you care about where your data sits, and whether you’re willing to pay roughly $20 a month.
What ChatGPT still does best
ChatGPT’s advantage isn’t raw intelligence. It’s surface area. Voice mode is good enough to use while cooking. Custom GPTs let a team package a workflow once and share it. Image generation, code execution, memory and project folders mean you rarely need to leave the app to finish something. If your work jumps between formats all day, that breadth beats a benchmark point or two.
The free tier is genuinely usable, which is why the user base runs into the hundreds of millions. The weak spots are consistency and personality. Default answers lean agreeable, and the model will sometimes nod along with a premise it ought to challenge. It’s also moving fast enough that opinions formed two years ago are worthless — the product has changed considerably in both directions, as this look at how ChatGPT has evolved makes clear.
One thing to watch: OpenAI has started testing advertising on the free and Go tiers in India, which tells you something about where the free product is heading. Details of that rollout are covered here. If you’re sensitive to ads appearing next to your work, budget for a paid tier or plan to switch.
Claude: the careful writer
Claude’s reputation rests on two things: long context and restraint. Feed it a 90-page PDF and it keeps the thread better than most competitors, and it tends to match a requested tone instead of flattening everything into the same cheerful register. For editing, rewriting and code review, that combination is hard to beat.
The trade-offs are real. There’s no native image generation, the ecosystem of third-party integrations is thinner, and heavy users hit usage limits faster than they expect. Pro costs the same $20 as ChatGPT Plus, so you’re choosing between polish and breadth rather than price.
Gemini: the one that lives inside everything
Gemini’s pitch is proximity. It sits inside Gmail, Docs, Android and Search, so asking it to summarise a thread or draft a reply is a two-click operation rather than a copy-paste exercise. The context window is enormous, and grounding against live search results is free and fast.
What you give up is coherence. Quality varies noticeably between the app, the workspace sidebar and the search box, and the sheer number of entry points makes it hard to know which version you’re talking to. For Google-heavy teams it’s often the pragmatic pick. For everyone else it’s the one that feels least like a single product.
Perplexity and the search-first tools
Perplexity answers questions with citations attached, which changes how you use it. You’re not asking for prose, you’re asking for a defensible answer with links you can check. For market research, fact-checking and quick comparisons, that’s faster and more trustworthy than a chatbot confidently paraphrasing its training data.
Ask it to write something long and transformative, though, and the seams show. It summarises well and reasons less well, and the writing has a clipped, report-like quality. Treat it as a research instrument rather than a writing partner.
Open-weight models: control, at a cost
Llama, Mistral and similar open-weight releases can run on your own hardware or a rented GPU. Nothing leaves your network, there’s no per-seat licence, and you can fine-tune on your own material. For regulated industries, that’s often the only option that survives a procurement review.
The bill arrives in engineering hours. Setup, prompt tuning, updates and uptime become your problem, and the quality gap on genuinely hard tasks is still noticeable. It’s a strong fit if data control is non-negotiable and you have someone technical around. It’s a poor fit for a two-person marketing team.
The money and the fine print
Consumer pricing has settled into a narrow band: ChatGPT Plus, Claude Pro, Perplexity Pro and Google’s AI tier all land near $20 a month, with annual discounts. API pricing is a different world, billed per million tokens, and it can be dramatically cheaper or dramatically more expensive depending on how you use it. Chat interfaces are subsidised; programmatic access is not.
Regulation is the other variable. The EU has been pushing ChatGPT and similar tools toward stricter transparency and risk obligations, a process summarised here, and government buyers are already building separate, locked-down deployments rather than relying on consumer accounts — the Pentagon’s own ChatGPT and Grok setup being the obvious example. What’s permitted in a consumer app and what’s permitted in a contract are drifting apart.
Where the differences really bite
Every model hallucinates, and every model flatters. The failure modes just look different. Some invent citations that read perfectly. Some agree with you until you change your mind, then agree with that too. There are documented cases of assistants reinforcing a user’s worst interpretation of reality rather than questioning it, including a widely discussed exchange with a man who said he felt delusional. Pick the model that fails least badly in your specific use case, not the one at the top of a leaderboard.
A practical way to choose
- You write for a living: start with Claude, keep ChatGPT for images and voice.
- You research for a living: Perplexity for sourcing, then paste the material into a chat model to synthesise.
- You live in Google Workspace: Gemini, mostly for the proximity.
- You handle sensitive data: open-weight models you host yourself, and accept the maintenance overhead.
- You want one subscription and no fuss: ChatGPT Plus remains the safest single pick.
The 20-minute test that settles it
Before you commit a year of your budget, run the same two tasks through each free tier. First, paste a 1,500-word document you know intimately and ask for a summary — you’ll spot invented details instantly. Second, ask it to rewrite a paragraph in your voice and read the result aloud. Whichever model passes both without embarrassing itself is your answer.
Then spend an hour learning to prompt it properly. A well-specified request on a mid-tier model routinely beats a vague one on the best model available, and the difference is bigger than anything the vendors advertise. If you want a structured approach to that, this walkthrough on getting genuinely good answers out of ChatGPT is worth the read, and most of it transfers to whichever tool you end up choosing.