I Compared AI Subscription Plans. So You Don't Have To

Most people don’t pick an AI subscription because they love comparing tiers. They pick one because something broke at the wrong time. A deadline. A messy doc. A question that needs sources. A code change that shouldn’t ship until it’s actually understood. It’s the moment you’re staring at a long paste, thinking, “I just need this to work for the next two hours.”

At roughly $20 a month, the top assistants are closer than the ads make them feel. The separation shows up when you care about a specific workflow or when your usage gets heavy enough that limits start shaping what you can finish.

I’m keeping this simple: what each plan says you get, and the practical differences you actually notice.

OpenAI ChatGPT

ChatGPT isn’t only a chat tab anymore. It’s a bundle that includes Codex as a separate coding surface and Sora as a generation product, each with its own limits and queue behavior.

We’re in the GPT-5.2 cycle now. The difference most people feel isn’t a sudden leap in brilliance. It’s how often you get forced to slow down or stop.

Codex is best thought of as a separate macOS workspace for delegation. A concrete “this is why I use it” moment looks like handing it a repo task (refactor a module, add tests, prep a PR), letting parallel agents run, then reviewing the diff like you would in a normal code review.

Sora has a similar “you feel it when it’s busy” reality. A common experience is starting a render and watching it sit in queue when demand spikes. Higher tiers generally get treated as higher priority, but peak periods can still drag.

Anthropic Claude

Claude’s ladder is simple. Pro is the paid baseline, and Max mainly buys you room to work without interruptions.

A grounded example here is the “four-pass rewrite.” You paste a long brief, ask for structure, then tighten tone, then add missing points, then compress it. Pro can handle that, but Max is where you stop worrying about whether the fourth pass is going to get interrupted.

Cowork fits into the paid story, but the felt difference still comes back to one thing: how long you can keep going before the tool taps out.

Google Gemini

Gemini’s paid tiers can look like a bundle, but the detail that changes expectations is that creative output is metered.

Right now we are at Gemini 3 and a concrete way people trip on AI Plus is treating it like an unlimited image studio. You do a big batch of variants for one project, then another batch the next day, and suddenly the credits are the main character. Used as a “creative burst” tier, it feels like a steal. Used as a daily factory, it gets frustrating fast.

Perplexity

Perplexity is the most research-first tool in this group. The upgrade question is about volume and workflow, not chat personality.

The most “this is why I pay” example is when your day includes “search my stuff.” You connect Drive, then ask for a specific line or claim from a document you wrote weeks ago, and you want the answer to come back with the exact supporting snippet. That’s where connector coverage stops being a bullet point and starts being the reason the subscription exists.

The simplest way to read this market

At around $20, you’re mostly picking a daily default: the interface you like and the jobs you repeat. Paying for two $20 plans is a waste when they solve the same job in the same way.

Where subscriptions stop overlapping is outside the chat box: Codex as an agent workspace, Claude as a long-session drafting loop, Gemini as a credit-metered creative layer, Perplexity as a research-and-connectors workflow. If a second plan buys you a different environment you’ll use every week, keep it. If it mostly buys you another place to type prompts, it usually isn’t worth it.