How I Pick AI Subscriptions by Budget and Role

When people ask which AI plan to buy, they usually start with budget. I start somewhere else: will most of your work be developer work, or not?

Same monthly spend, completely different stack. Below is the map I use after paying for and living with these plans myself. Five options for non-builders, five for builders.

Non-builders

Google AI Plus is the lightest on-ramp. Around ₩7,500/month gets you image and video generation plus Drive integration. It fits people who want AI woven into daily life more than deep work automation.

ChatGPT Plus is my default in the ₩30,000 band. Documents, light automation, and a soft entry into vibe coding all land here first. If you only keep one paid plan in this range, this is the one I would keep.

Gemini Pro sits in the same ₩30,000 band, but only if you already live inside Google Workspace and you do not need a strong agent loop. It feels natural when Docs, Gmail, and Drive are already your desk.

Cursor Pro is the first real choice when a non-builder wants to vibe code. It is best for coding sessions, not for everything else. Outside those sessions I still prefer ChatGPT or Claude as the daily chat. Devin Pro is another option in this band if you specifically want to try agent-style product building.

ChatGPT Pro 5x or Claude Max 5x is the heavy-content band, starting around ₩150,000/month. For image volume and raw token headroom, ChatGPT Pro 5x wins. If writing is the center of the job, Claude Max is the better fit.

Builders

OpenCode plus OpenRouter free APIs is the zero-won starting point. You can touch a wide range of models without a card on file, and the CLI path is natural for developers.

OpenCode Go is useful if you want to experiment with open-source models for about ₩15,000/month. Expensive open models still do not stretch far on this plan, so treat it as a sandbox, not a production budget.

Grok SuperGrok is my value pick for vibe coding around ₩45,000/month. Token volume and generation speed are the reasons. Model updates on this side of the market keep moving, which matters when you live in the loop every day. Kimi-class models can look attractive on paper, but in practice I have not found them cheap once real agent traffic starts. For pure cost-performance I still land on ChatGPT and Grok first.

GLM-5.2 is a strong model that I would rather consume through a product that already packages it than through a bare subscription. The raw plan has been weak on speed and token limits relative to how useful the model is.

ChatGPT Pro 20x plus Claude Max is the production band. Around ₩300,000/month, ChatGPT Pro 20x alone is the first stop. From about ₩450,000/month I move to Claude Max 5x plus ChatGPT Pro 20x.

The expensive models still win, with a catch

Claude’s Fable 5 still looks like the best model I have used, including after spending time with GPT-5.6 Sol. The hard part is price. Opus-class usage only makes sense when you can point the spend at work that returns money. That is also why ChatGPT remains a strong product even when it is not the peak model: the economics are easier to live with.

For teammates who are expected to produce real output with AI, I now treat roughly ₩600,000/month as a serious baseline. In the US, weekly spend above that per person is not rare in aggressive teams. The open question is whether the work actually earns it back. Spreading working patterns across the team matters as much as buying the top plan.

I will keep updating this map as plans and models move. The first filter stays the same: builder or not, then budget.