At the index’s default assumptions, Claude Pro shows a 12.2× API-equivalent multiple. MiniMax Plus — same $20/month — shows 7×. Read those as a ranking and you buy the wrong plan, because MiniMax Plus delivers tokens at $0.051/M against Claude Pro’s $0.513/M: exactly 10× cheaper per token, with the smaller “×” next to its name. The multiple is not a discount score. It tells you how far the subscription beats the *same provider’s* API list price — a switch-to-API signal, never a cross-provider value ranking.
What the multiple actually computes
The API-equivalent multiple is your month of estimated tokens priced at the provider’s own API list rate, divided by the subscription price. The denominator of value is the plan; the baseline is whatever the provider chooses to charge API customers — and that baseline varies enormously.
For Claude plans the baseline is Claude Opus 4.8 API pricing — $5/M input, $25/M output, which our fixed 15:1 input/output blend turns into $6.25 per million tokens (per Anthropic’s pricing docs). For MiniMax plans the baseline is MiniMax-M3 at $0.30/M in, $1.20/M out — a $0.36/M blend. Divide anything by a baseline 17× larger and the quotient looks 17× more impressive. That is the whole trick.
The head-to-head, computed from the live model
Run both $20 plans through the index at its defaults (25,000 tokens per interaction, 40 active hours per week). Claude Pro’s 45-messages-per-5-hours works out to an estimated 39M tokens a month — $0.513 per million, worth $243.75 at Opus API list, a 12.19× multiple. MiniMax Plus’s 4,500 requests per 5 hours works out to roughly 390M tokens — $0.051 per million, worth $138.96 at M3 API list, a 6.95× multiple.
Same sticker price, ten times the tokens per dollar — and the smaller multiple sits on the better deal. If the multiple were a value ranking, that would be impossible.
At the extremes, the two rankings barely speak
Across all 64 plans in the live index, the two highest multiples in the entire dataset belong to plans that are nowhere near the cheapest per token — and the cheapest plan of all sits mid-table on the multiple.
| Plan | Price | Effective $/M | API multiple | Rank by $/M | Rank by multiple |
|---|---|---|---|---|---|
| ChatGPT Pro 20× | $200/mo | $0.256 | 25.6× | 9th–10th | 1st |
| Claude Max 20× | $200/mo | $0.256 | 24.4× | 9th–10th | 2nd |
| Claude Pro | $20/mo | $0.513 | 12.2× | — | — |
| MiniMax Max | $50/mo | $0.038 | 9.3× | 1st | 10th |
| MiniMax Ultra | $120/mo | $0.046 | 7.7× | 2nd | 16th |
| MiniMax Plus | $20/mo | $0.051 | 7.0× | — | — |
ChatGPT Pro 20× and Claude Max 20× top the multiple ranking at 25.6× and 24.4× — while ranking only 9th and 10th cheapest per token. MiniMax Max, the single cheapest plan in the dataset at $0.038/M, ranks 10th on the multiple. Sort one way and you get the American frontier labs; sort the other and you get the Chinese ones. The metric you pick decides the winner before the data does.
The mechanism: the baseline varies more than the plans do
The multiple’s baseline — the provider’s blended API list price — spans more than a 17× range across the dataset: Opus 4.8 blends to $6.25/M and GPT-5.5 to $6.56/M, while MiniMax-M3 blends to $0.36/M and DeepSeek V4-Pro to $0.46/M. A subscription sitting on an expensive API inherits a structurally large multiple regardless of how generous the plan itself is. The multiple co-measures two things at once: the deal, and the provider’s own list-price ambition.
It is also acutely assumption-sensitive. DX’s 2026 pricing analysis runs the same subscription-vs-API comparison for Claude Max and lands at roughly 3–7× — not our 12–24× — because it assumes a different usage shape. Neither figure is wrong; both are functions of their inputs. That is one more reason a single “×” should never be read as a fixed property of a plan.
How to read the multiple — and how the index already does
Use effective $/M to rank plans against each other; use the multiple only to answer one narrower question: *should I subscribe, or pay this same provider’s API rates instead?* Above 1×, the subscription beats the API for your usage shape. Below 1×, pay per token. That question the multiple answers precisely — it just answers nothing about the provider next door.
This is already how the index is built, not a correction we are announcing. The table’s default sort is effective $/M, and the “Best value” verdict is defined as the lowest $/M — which under current defaults surfaces MiniMax Max, not the 25× plans. Sorting by the multiple is there too, one header-click away, because the switch-to-API question is worth answering. It is labeled a comparison against API list price. Now you know exactly why.