What we could not confirm, and why that matters

A deliberately unusual thing to put on a website

Scattered across this site's model pages and reference tables are a handful of plain admissions: a specific figure we could not confirm against a source we treat as authoritative, stated directly rather than papered over with a plausible-sounding estimate. That's an unusual choice for a site whose whole premise is being useful for cost decisions — most sites in this position would either quietly guess or quietly omit the topic entirely, rather than naming the gap out loud.

Why guessing is the worse failure, even though it's less visible

A wrong number that looks confident is more dangerous than an admitted gap, because nothing about it signals caution. A reader budgeting against a plausible-but-wrong figure has no way to know it needs independent verification — it looks exactly like every other number on the page. An admitted gap, by contrast, tells you precisely where to be careful, which is strictly more useful than a number that might be silently wrong in either direction.

Where these gaps actually come from

Most of the figures we couldn't confirm aren't obscure trivia — they're details that simply weren't restated in the specific source this site treats as authoritative for a given fact, even though the broader claim around them is well-supported. A legacy model's exact context window, a specific fast-mode rate on a model where fast mode's existence is confirmed but its price isn't, whether a retirement date that's already passed actually executed on schedule — these are narrow, specific gaps, not sweeping uncertainty about the whole lineup.

The discipline this requires, and why it's worth the friction

Every figure on this site traces back to a specific source, checked on a specific date, and the build itself enforces that discipline — a fact that goes too long without being re-verified fails the site's own freshness check rather than sitting there quietly aging into a lie. Admitting a gap is the natural extension of that same discipline: if the standard is "verified against a real source, on a specific date," then anything that doesn't clear that bar has exactly one honest option, which is saying so.

What this means for how you should read any number here

Treat the absence of a "could not confirm" note as a real signal, not a formality — it means the specific figure was checked against a source this site treats as authoritative, on the date shown. Where a note is present, that's not a bug in the page; it's the page telling you precisely where its own confidence ends, so you know where to do your own independent check before relying on it for something that actually matters.

Why this approach doesn't scale to "confirm everything eventually"

It might seem like the goal should be closing every one of these gaps over time until none remain, and for gaps that resolve through better sourcing, that's exactly right. But some of them are genuinely structural — a legacy model's exact specs that were never restated anywhere current because the model itself is being phased out, not actively maintained. Forcing a confident-sounding number into that kind of gap just to make the page look complete would be exactly the failure mode this whole approach exists to avoid.

The broader point, beyond this one site

A number that's confidently wrong costs you more than a gap that's honestly admitted, because the wrong number doesn't prompt you to check it — it just quietly fails you later, at whatever moment you actually acted on it. That's true of any source you're relying on for a real decision, not just this one, and it's worth applying the same scepticism to anywhere else you're pulling a number from with real money riding on it being right.

What we've actually left unconfirmed, as of this writing

The current set is narrow and specific by design: a couple of legacy models' exact context windows and output ceilings, one current model's fast-mode rate where fast-mode support itself is confirmed but its specific price isn't, and whether a couple of already-passed retirement dates executed precisely on schedule. None of these undermine the site's core, actively-maintained pricing and capability data for the current lineup — they're specifically the oldest and most peripheral corners of what this site covers, which is exactly where you'd expect sourcing gaps to concentrate on any honestly-maintained reference.

See how we source every number on this site for the mechanics behind this discipline, and any model's reference page to see this principle applied directly, not just described.