「Enterprise」不等于贵，「Unknown」不等于没价：目录里的定价字段该怎么读

The short version first: a pricing label is the author's classification, not the vendor's

Every pricing tag in a directory is what somebody decided after reading the vendor's page. Done well it saves you three minutes; done badly it gives you an hour of confident illusion. This article dissects our own fields, because all 376 records live in the repository and every claim here can be re-checked line by line.

1. The real distribution of our five values (and what they do not mean)

pricing.model is restricted to five values. Measured on 2026-09-18:

Three things in that table contradict the usual reading:

1. unknown does not mean 'the vendor publishes no price'. 29 of the 81 do carry digits — github-copilot is '$10/mo / $19/user/mo', midjourney is '$10-$80/mo'. They landed in unknown because no single tier can be circled: ranges, add-on prices and per-seat ladders do not fit freemium, enterprise or usage.
2. freemium does not mean 'cheap'. It is the biggest bucket at 247, and it holds both 'Free / Pro $9/mo' and 'Free / Business from $99/mo'. The tag answers 'is there a free entrance', never 'at what magnitude'.
3. The free bucket is the emptiest and the one to distrust. All ten contain no digit — a record that says 'free' without a quota has said nothing.

2. The only readable thing is the copied line, and it is short

pricing.detail.zh is the single field where we copy the vendor's wording. Counting it on the same day: median length 16 characters, longest 96, average 18. We keep it verbatim — 'Free / Plus $20/mo / Pro $200/mo', '$10/user/mo (Notion add-on)', 'billed per second / optional Pro plan'. No beautifying, no converting: the moment we convert, we have inserted an assumption about your volume.

3. Four formats that look comparable and are not

Normalize each into one comparable sentence: multiply per-seat price × seats; for 'from' prices first ask whether the advertised tier includes what you need; for usage-based products multiply your real monthly volume (that is exactly what /free-tools/llm-api-cost-calculator does — it only accepts numbers you provide); for bundled products charge the cost to the host subscription instead of pricing it standalone.

4. What to ask when 45 records say 'contact sales'

45 of the 81 unknown records read 'contact us / enterprise pricing TBD / waitlist', as do 19 of the 37 enterprise ones. Those are not 'unfindable', they are 'not decided by a page'. To get comparable numbers, ask four questions:

1. Minimum contract term and minimum seats — these decide what you actually pay per year; pages usually show only the unit rate.
2. The unit price after the included volume — the quota and the overage rate often come from two different documents.
3. Whether admin and security seats are billed separately — some charge per team, some per person.
4. Whether data retention, private deployment and audit logs cost extra — that is where enterprise tiers really diverge.

Write the answers into one table. Do not read 'quote only' as 'expensive', and do not treat a first-year discount as a standing price.

5. What these fields cannot do (stated plainly)

• The labels are ours and can be wrong. Another reader may reclassify a record from unknown to usage. That is why we print the copied line beside the label: you can overrule us.
• We do not quote, do not negotiate on your behalf, and do not accept vendor-supplied 'exclusive discounts'; where a link is an affiliate link, the page says so.
• Numbers expire: each record carries its own lastVerifiedAt, and the vendor's pricing page governs on the day you buy.

One-sentence summary

Read the copied line first, the label second. The label answers 'which bucket', only the line answers 'what you pay'. Anyone who treats the bucket as the conclusion is budgeting with somebody else's filing system.