Your1M.com User Guide
Your1M in One Sentence
When you're torn between multiple channels and selling methods — "which one is actually cheaper?" — Your1M uses your Usage Pattern as the ruler, converting very different AI offers into a comparable effective cost per Your1M.
Your1M represents 1,000,000 Input + Final Response tokens under your Usage Pattern. Thinking does not increase the Your1M denominator, but any Thinking cost charged by the provider is included in the calculation.
We Compare Offers, Not "List Prices" and "Sale Prices"
Terms such as "list price," "sale price," and "discount" are marketing language. They do not always accurately describe the transaction a user can actually make. Some prices are available only for a first purchase, some only for renewal, and others depend on identity, region, commitment period, or other eligibility requirements. Some products may also remain "on sale" for long periods while the advertised "list price" is rarely—or never—the price users actually pay.
For this reason, Your1M does not decide which price should be considered the "real" list price or which one should be considered promotional. Different transaction conditions are treated as different offers.
For example, the same plan might be available for ¥299.9 to users eligible for a first purchase, ¥360 to users eligible for a first renewal, and ¥600 to users without either eligibility. To Your1M, these are not one "regular price" and two "discount prices." They are three real offers with different eligibility conditions.
What matters is: whether you're eligible for the offer, how much you actually pay, and what you actually get.
Each offer is evaluated independently according to its actual price, included capacity or service, eligibility requirements, validity conditions, and usage restrictions. When those conditions differ, the offers can coexist as separate ways to buy the same product.
The goal is to strip away marketing labels such as "list price," "X% off," and "limited-time deal," and compare the transactions users can actually make: under what conditions, how much do you pay, and what do you get?
On long-term subscriptions: AI models change quickly, and long commitments reduce how
comparable different offers really are. So we generally exclude discounts that require a
commitment longer than about one month — unless the difference from a monthly offer is large
enough to significantly change a purchase decision, which we evaluate separately. What matters is
the shortest non-cancellable commitment you take on to complete the purchase, not whether the
vendor writes /month or /year.
Core Concepts
Usage Pattern: the shape of "how you use AI" — token composition (input/output ratio), cache-hit habits, context-length distribution, weekly time distribution, and execution flexibility. A profile is not your identity; one profile corresponds to one usage scenario.
Your1M: the unified measure — 1,000,000 Input + final Response tokens under your Usage Pattern. Different offers are all converted onto this one scale so they can be compared fairly. The effective cost per Your1M is therefore the price of that one million-token unit, not a prediction of your bill.
We estimate reasonably, based on your usage pattern. For each offer we work out, under your usage pattern, how much capacity it can yield and what it costs. For prepaid offers this means the maximum usable capacity — how much Your1M that offer could yield if you used all of its usable resources on this one model. It is not a prediction of what you will actually spend; it is the offer's economic capacity under one consistent set of conditions. That is why we say effective cost rather than actual cost.
Thinking tokens: they do not increase the Your1M denominator — the denominator is the workload you can understand and describe (input + final response). But if a provider charges for thinking, that cost is included in the calculation.
Which Models We Track
AI models evolve quickly. Your1M does not aim to list every model. Instead, we maintain a curated model universe focused on the models people actually compare across multiple providers.
Why not cover everything? The value of comparison comes from accuracy, not volume. To compare a model fairly, we need to break its offers down to the parameter level and continuously track changes in pricing, capacity, and rules. The broader the coverage, the harder it becomes to keep every entry reliable.
Our general criteria: We focus on large language models from major developers that are currently sold, meet a meaningful capability threshold, are widely purchased, are impractical to deploy on consumer-grade hardware, and have usage that can be converted into Token-based pricing.
How we handle model versions: Some models are released or updated multiple times. When an upgrade is broadly adopted across providers while the API identity remains continuous, we generally treat it as the same evolving model. This may include Preview-to-stable releases, continuously updated models, or post-trained revisions. In these cases, the release date reflects the most recent update.
If providers continue serving an older version after a newer one is released, and the two remain separately callable, we treat them as different models—for example, version 2.7 and version 3.8.
Not listed ≠ not considered. We also maintain exclusion records. When a model has been evaluated and we decide not to include it for now, that decision is recorded so the same question does not need to be researched repeatedly. If you cannot find a model in Your1M, it may be not yet evaluated, or evaluated but not currently included.
Don't see your model?
Use "Report error" to tell us which model you use and where you buy it. It will go directly into our evaluation queue.
Which Providers We Track
The same model is often available through many providers: directly from the model developer, through cloud and compute platforms, through aggregators, or even through different regional offerings from the same company.
Your1M tracks these providers, but what we actually compare are the offers they provide—not the providers themselves.
Which providers do we focus on? We generally track providers that have meaningful market adoption or attract enough attention that people regularly discuss whether their offers are worth buying. The purpose of tracking a provider is to discover and maintain offers worth comparing, not to evaluate the provider as a whole.
Providers themselves are not comparison entries. What enters the comparison is always a specific offer from a specific provider. A provider is within our scope because it has offers worth tracking. A provider may be excluded—or later leave the scope—simply because it no longer has offers worth comparing separately, or because what it provides is already sufficiently covered elsewhere.
A single provider may offer pay-as-you-go pricing, batch pricing, subscriptions, credit packs, and other ways to buy the same model. We include and calculate each offer separately rather than assigning a single generalized price judgment to the provider.
Once a provider enters our tracking scope, we also review it periodically for newly introduced offers.
Common reasons a provider may not be included:
- Insufficient comparability: Public and verifiable evidence for pricing, capacity, capabilities, or rules is not sufficient to reliably normalize its offers.
- Insufficient relevance: Market usage and discussion are limited enough that continuously maintaining the provider adds little comparison value.
- Duplicate coverage: Its offers are effectively already represented elsewhere and add little new information to the comparison.
Not listed ≠ not considered. Provider evaluations are recorded as well. If we have reviewed a provider and decided not to include it for now, we keep the reason so the same question does not need to be researched repeatedly.
Don't see a provider you use?
Create an account and talk to the Assistant. Tell us which provider you use and why you think it belongs in the comparison. The Assistant can submit it directly for evaluation.
Quick Start
- Build a profile: click "Profile", pick a scenario template or generate one quickly via the questionnaire, or fine-tune manually (weekly distribution, context density, token composition, execution flexibility).
- Choose models: check the models you want to compare in the left rail (unselected models won't appear in the table).
- Compare: the table lists each "offer × model" pair's "Your1M" unit price and capacity; click a column header to sort.
- See details: click a row to expand the calculation explanation on the right — Evaluation scope → Calculation steps → Adopted parameters → Parameter evidence. The evaluation scope first states which profile × model × offer this result is computed for; the calculation steps give each formula line with its result (recomputable down to the final Your1M unit price); parameters with an adopted evidence reference carry a
[1]-style citation, and the evidence section lists the source (or a supporting source), the observation time, and a link to the original page; parameters without evidence are simply not annotated.
On sources: we prefer precise official information when it is available. When it is not, Your1M may use reproducible third-party measurements, cross-checked against other evidence. What matters is that every figure is traceable and verifiable — not that it comes from one particular kind of source.
Sorted by Your1M, not overall service quality. The table ranks offers purely by their Your1M economics. It does not rank channels or providers overall — support quality, ecosystem, reliability and other factors are real, but they are not part of this number.
Selling Method Families
A Selling Method Family groups offers that share the same underlying economic structure and the same calculation template — so they can be explained and compared in the same way.
| Family | Meaning |
|---|---|
| Plan pool | A fixed prepaid pool, allowance amortized into a Your1M unit price |
| Pay-as-you-go | Pay for what you use, no capacity cap |
| Batch | Async batch, with a batch discount |
| Tiered | Priced by context-length / usage tier |
| Agent pack | An allowance billed by agent capability / invocation |
| Not comparable | No published token conversion — excluded from unit-price comparison |
"Not comparable" is a judgment about content, not about the label: a membership, for example, participates normally when it includes a clearly stated AI entitlement that can be reliably converted; it is marked not comparable only when its AI entitlement cannot be reliably separated from the bundle (e.g. sold together with cloud storage and search).
Mechanism features (e.g. off-peak discount, 1:1 deduct, cache-aware pricing, validity-limited, credit allowance, membership) appear as badges at the top of the detail view; hover for an explanation. These are additional characteristics that do not change the core calculation — they refine a family rather than forming a new one.
Account and Features
- Save profiles: after signing in, save your tuned profiles to the library and reuse them with one click.
- Report an error: spot a wrong price or calculation? Click "Report error"; the assistant will collect your feedback and route it into the processing pipeline.
- Assistant: ask anytime from the bottom-right corner, e.g. "Why is this offer more expensive? Which one is cheapest?".
- Language: switch between 中文 / English in the top-right corner.
Important Note
The prices, capacities, and calculation results shown by this service are for reference only and do not constitute purchase advice. Model prices are published by suppliers and may change at any time; if you find an error, please report it via the correction feature. The final purchase decision and transaction are between you and the supplier.