Fastino GLiDE Decision Index is the lab’s own run, repeated once

At 1 Oct 2026, 17:21 ET, PR Newswire carried a Fastino Labs release, and that release is the whole evidence base for Fastino GLiDE Decision Index. Fastino Labs, in San Francisco, released GLiDE, model ID fastino/GLiDE. On Fastino’s own run of Decision Index 0.2.1, covering 155,390 requests, GLiDE scored 64.81 versus Jev’s published 57.91, a 6.90 point gap. The release says it led all five areas and 31 of 38 benchmarks. Scores in this article are that lab run. They are not an outside re-score.

TL;DR

What actually happened

One lab published a score, and a second site printed it again. That is a thinner stack than a leaderboard headline suggests. The primary text is the PR Newswire release. The time on it is 1 Oct 2026, 17:21 ET. The company is Fastino Labs, placed in San Francisco. The model name in the release is GLiDE. The model ID is fastino/GLiDE.

The index name and version matter. Decision Index 0.2.1 is what they ran. The request count on that run is 155,390. The headline comparison is GLiDE at 64.81 against Jev’s published 57.91. The gap they state is 6.90 points. “Jev’s published” means the release is comparing against a published number, not claiming Jev re-ran the test for this announcement. This page does not add a biography of Jev or a second lab’s confirmation.

“Led all five areas and 31 of 38 benchmarks” is the release’s breadth claim. The facts used here name some of those areas and two benchmarks. They do not name the other areas or the other benchmarks. This article will not fill the missing titles in.

The CEO line is separate from the index. George Hurn-Maloney said GLiNER has passed 50 million downloads. That sentence is about GLiNER, the name he used, not a second score for GLiDE. Downloads are not Decision Index points. Keep them in different sentences.

The wires are PR Newswire and Third News. Third News repeats those scores. It does not, in this assignment, count as a separate evaluation.

What the lab scored on its own run

Treat every point below as Fastino’s own run. If a friend says “the index found,” correct it to “Fastino reported from its run of the index.”

Knowledge and Reasoning: 62.9 versus 51.4, a stated gap of +11.5. Tools and Automation: 83.5 versus 75.1, a stated gap of +8.4. Arts and Human Taste is given as +8.3. The two base scores for that area are not in the facts used here, so they are not invented to make the row look complete. A plus without the pair is what the source gave.

CRUXEval accuracy: 92.6% versus 73.0%. CLadder: 88.7% versus 72.6%. Those are the two named benchmark accuracies. The “31 of 38” line says the lead was wider than these two names. It does not authorize a list of the benchmarks that were not named here.

The context window is 40,000 tokens. That is a capacity figure in the release, not a score on the index. About two-thirds of test requests stayed on the fast path. About two-thirds is the release’s hedge. This page does not convert it into a precise request count.

The 6.90 point gap is the overall gap they state between 64.81 and 57.91. The area gaps are additional, not a second overall. +11.5, +8.4, and +8.3 describe slices. Do not add those slices into a new total.

Fastino GLiDE Decision Index is a fair name for this packet because the model, the lab, and the index version travel together. It is an unfair name if “index” is heard as “independently administered.” The run is the lab’s.

Why the second page is not a second eval

Third News repeats those scores. Read that verb narrowly. Repeating a release is how a number shows up on a second URL without a second experiment. The second page reads as a rewrite of the release, not a separate eval. Two links, one measurement.

A rewrite can still be useful. It tells you the score package traveled. It does not tell you a different team loaded fastino/GLiDE and called Decision Index 0.2.1 on 155,390 requests of its own. If the rewrite matches, the match is expected. Agreement with a rewrite is not corroboration.

So the confirmation count a careful reader should feel is smaller than a pair of headlines. The scores to cite are the lab’s own run, via the release, with Third News as an echo. Echoes do not move 64.81 from “announced” to “replicated.”

No prediction market sits on this result in the facts used here. There is no Yes price on whether a later run still leads by 6.90. Do not borrow odds language. A 6.90 point gap on the lab’s run is not a probability.

The download sentence stays an echo too, if Third News repeats it with the scores. It is still George Hurn-Maloney’s statement that GLiNER passed 50 million downloads. An echo does not audit a download counter.

What to do as a reader (not a trade)

Quote the score with the owner of the run in the same breath. Fastino’s own run of Decision Index 0.2.1, 155,390 requests, GLiDE 64.81, Jev’s published 57.91, gap 6.90. Then the slices you actually have, and not the slices you wish the release had printed in full. Arts and Human Taste stays “+8.3” unless you are looking at a page that shows the bases.

Say the second page is a rewrite. If someone cites Third News as an independent leaderboard, send them back to the wording. Repeating a press release is a distribution fact. It is not a replication.

This is not a trade and not a model-shopping order. Nothing here says to switch vendors, buy access, or expect the gap to hold on your tasks. A context window of 40,000 tokens is a spec line from the release, not a promise about your documents. About two-thirds on the fast path is a statement about their test requests, not about your traffic.

GLiNER’s 50 million downloads, as stated by the CEO, do not raise or lower the 64.81. Popularity of a related named project is a separate brag in the same release. Let it stay separate so it cannot smuggle authority into the index.

Other site notes are not evals of this model. The DogeOS public testnet is unrelated. The Gemini 4 Argon launch is unrelated. Do not stack them under Fastino to make a trend.

A clean card for Fastino GLiDE Decision Index lists the ET timestamp, San Francisco, Fastino Labs, the ID fastino/GLiDE, index version 0.2.1, request count 155,390, the 64.81 / 57.91 / 6.90 trio, the five-area and 31-of-38 claims, the named slices including the incomplete Arts line, CRUXEval, CLadder, 40,000 tokens, about two-thirds on the fast path, and the CEO’s GLiNER download line. Then it says: lab’s own run; second page is a rewrite, not a separate eval. That last clause is the whole skepticism. You do not need a harsher one, and you should not invent a counter-score.

Keep Fastino GLiDE Decision Index tied to the lab that ran it. The release time stays 1 Oct 2026, 17:21 ET, because that is the clock on the wire. If you translate the zone later, do not drop the original. The number people will argue about is 6.90. The thing they should argue about first is who ran the test.