Skip to content

Editorial policy

Every specification and price on this site is transcribed from a primary source and carries the date we checked it. When a figure has not been published we say so rather than estimating, and when we get something wrong we correct it visibly.

Effective

Our principles

This site exists because most coverage of AI models is either vendor marketing restated, or numbers copied from other blogs that copied them from somewhere else. We try to do something more boring and more useful.

  • Primary sources or nothing. If we cannot trace it, we do not publish it.
  • Say the unflattering thing. If a model is expensive, slow, or worse than a rival at something, we write that.
  • Unknown is a publishable answer. A blank with an explanation beats a confident guess.
  • Date everything. A price without a date is not information, it is a rumour.

How we source facts

Model specifications, context windows, pricing, release dates and retirement dates come from the vendor’s own documentation. For Kimi models that means:

All model data lives in a single structured file in our codebase, and every page reads from it. That means a price appears in exactly one place: when Moonshot changes it, we change one line and every page, table and calculator updates together. No page can quietly go stale while another is correct.

Each entry records the date it was last checked against the source, and that date is what you see as “last updated”.

What we do when we do not know

Vendors do not publish everything. Moonshot does not state a maximum output token count for every model, or a release date for every variant.

When that happens we render “Not published” or a dash. We do not estimate, interpolate from a previous model, or repeat a number we saw on another site.

Why everything carries a date

AI model pricing and availability change on a timescale of weeks. Moonshot retired four model lines in under a year. A page that says “Kimi K2 costs $X” with no date is actively misleading once that model is discontinued.

So every content page shows when it was last verified, and every table cites its source.

This also means you should check the date. If it is months old and you are about to spend money, click through to the source we link and confirm.

How we use AI

We think readers deserve a straight answer on this, since we write about AI.

We use AI tools as part of producing this site — for drafting, editing and building the software. What we do not do is publish unreviewed model output. Every factual claim is checked against a primary source by a person before it goes live, and the structure, judgements and opinions are ours.

That is the line we hold: AI can help write a sentence, but it cannot be the source of a fact. Models confidently state pricing that is a year out of date, and that failure mode is exactly what this site is meant to protect you from.

Independence and money

We have no relationship with Moonshot AI. Nobody pays us to cover a model, to cover it favourably, or to leave one out. See our disclaimer for the full position.

We intend to fund this site through advertising. If and when that happens:

  • Ads will be clearly distinguishable from editorial content
  • No advertiser will get a say in what we write or how we rank a model
  • Ads will never appear above the main heading or inside navigation
  • They will load only after you consent, as described in our cookie policy

If we ever publish sponsored content or use affiliate links, we will label it clearly on the page itself. As of the effective date above, we do neither.

Corrections

We will get things wrong. When we do, we want to know, and we would rather hear it from you than leave it up.

How we handle it:

  • Factual errors are fixed as soon as we can verify the correct figure
  • The “last updated” date changes when we make a substantive correction
  • Where a correction materially changes a recommendation, we note what changed rather than silently editing

Spotted something wrong? Email [email protected] — ideally with a link to the source. We treat corrections as the highest-priority mail we get.

Report a correction →

What we do not do

Being clear about our limits is part of being trustworthy.

  • We do not run our own benchmarks. Independently benchmarking frontier models requires resources we do not have. Where we cite benchmark results we attribute them to whoever produced them, and we do not present other people’s numbers as our own testing.
  • We do not have inside information. No early access, no briefings, no leaks. Everything here is public information, researched carefully.
  • We do not provide support for Kimi products. We cannot help with your account, billing or API keys.
  • We do not give professional advice. We publish reference material. What you do with it is your decision.

Questions about this page? Email [email protected]. We are an independent publisher, not a law firm — nothing here is legal advice.