Volume changes everything. At ten items a day, model choice is a rounding error. At 50,000 a month, the difference between the flagship and the small model is the difference between a real line on the accounts and a coffee.
Scenario 5: classification at scale on GPT-5.5
A small online retailer tags every incoming support message as *refund*, *delivery*, *product* or *other*. About 50,000 messages a month, each roughly 300 input tokens, with a one-word answer of about 10 output tokens. They set it up on GPT-5.5 ($5 in / $30 out per 1M) because it was the default in the dropdown.
Which would you trust?
Which is the cheapest model that still does this classification job?
Scenario 6: Perplexity Pro and ChatGPT Plus
| Task | Perplexity Pro $20 | ChatGPT Plus $20 | Overlap? |
|---|---|---|---|
| Web research with cited sources | Core feature | Search is built in | Yes |
| Summarising uploaded reports | Yes | Yes | Yes |
| Drafting the report itself | Adequate | Strong | Mostly |
| Access to several vendors' models | Yes, in one place | OpenAI only | No |
| Monthly total | $40 for one research workflow |
Verdict: keep one. Both search the web with citations and both summarise uploads. If Marcus's day is mostly search-then-cite, Perplexity Pro at $20 does it and ChatGPT Free covers the occasional draft. If his day is mostly drafting with some search, ChatGPT Plus at $20 does it and Perplexity's free tier covers the occasional deep lookup. Either way the saving is $20 a month, $240 a year, and the only genuine reason to keep both is a task one of them cannot do, which the table does not show.