Never state a number without its date. In a field that obsoletes claims monthly, an undated benchmark is misinformation with a delay timer built in.
I've started treating "as of" as one of the load-bearing phrases in anything I write about AI tools, because the alternative — a bare claim with no date attached — has a shelf life measured in weeks and gets read as if it has a shelf life measured in years. Somebody encounters the sentence eight months after I wrote it and has no way to know whether it was true when I wrote it and has since gone stale, or whether it was never true and I got it wrong. Those are very different failures and an undated claim collapses them into the same unreadable thing.
This field specifically obsoletes claims at a pace most fields don't. A model that was the best available option this month can be superseded within weeks. A price that was competitive can be undercut. A limitation I ran into and wrote up as a fact about how a tool behaves can be fixed in the next release and my sentence, still sitting there unedited, becomes actively wrong rather than merely outdated — it doesn't fade, it flips. Writing without a date on the claim isn't a stylistic omission in a field that moves like this. It's handing the reader a landmine with a fuse they can't see, because nothing in the sentence tells them the fuse exists.
The honest version of any claim like this has two parts, not one: the observation, and the moment the observation was made. "This model is the strongest option for this task" is not honest on its own. "This model was the strongest option for this task as of the date I'm writing this" is honest, because it tells the reader exactly how much weight to put on it and exactly what to go check if the decision they're making depends on it still being true. The date isn't hedging. It's the part of the claim that makes the rest of it verifiable at all.
I've caught myself, and had to go back and fix, sentences I wrote with total confidence that were true the week I wrote them and false three months later — a tool I described as lacking a feature that shipped the feature before I'd even published the piece. That's not a failure of research. It's the base rate for writing about a field that changes this fast, and the only defence against it isn't writing more carefully, it's writing so that the reader can tell, from the text itself, exactly how current the claim is meant to be.
The instinct to skip the date comes from wanting to sound authoritative, and a dated claim can feel like it's admitting the ground might shift under it. But the confident, undated version is the one that actually erodes trust once it's wrong, because a reader who catches you stating something false with total certainty stops trusting the next claim too, dated or not. A reader who catches an undated claim that's gone stale, with no date to explain why, reasonably assumes you either didn't know or didn't care that it would go stale. Neither is a good look, and both are avoidable with four extra words.
So the rule I actually follow now is mechanical rather than aspirational: no number, no benchmark, no "the best tool for this" without a date attached to it, in the sentence itself, not buried in a byline or a page footer nobody reads. Honest and dated beats confident and timeless, every time, in a field where timeless claims are just claims nobody has got around to falsifying yet, and the falsifying, in this field, happens faster than almost anywhere else I write about.