On September 30, the company Fastino launched GLiDE, a model that makes fast decisions for software and automations. It describes GLiDE as better than Jev, the TypeSafe model we covered earlier. That is true, with one important caveat: Fastino ran the tests itself. Here is what it means for a small business.
What GLiDE does, in short
Models like ChatGPT write text. GLiDE writes nothing: you give it a situation and a list of options, and it picks one and tells you how confident it is (Fastino). Example: an email from a customer and the options "complaint", "quote request", "invoice", "spam". The model picks one and tells you whether it is 95% sure or only 55%.
What is new compared with Jev is "adaptive thinking". When the decision is clear, it answers fast. When it is unsure, it takes more time to think (Fastino).
What the numbers say
On a test called Decision Index 0.2.1, GLiDE scores 64.81 and Jev 57.91 (Fastino). That is a gap of almost 7 points. GLiDE wins on 31 of 38 tasks, and on "knowledge and reasoning" it scores 62.9 against 51.4 (Fastino).
How solid is "better"
Be careful here. The tests and the opponents were chosen by Fastino, the company selling GLiDE (MarkTechPost). There is no independent check yet, and GLiDE is not on the public leaderboard because submissions are currently paused (Fastino). The same holds for Jev, whose numbers also come from its own maker.
Honest verdict: GLiDE looks like a good model worth watching, but "better than Jev" is for now a vendor claim, not a proven fact.
What matters more than a score
- Price. Jev costs 0.042 dollars per million input tokens, and outputs are free (MarkTechPost). GLiDE's price has not been made public (Fastino).
- Speed. Jev answers in 70 to 500 milliseconds (MarkTechPost). There are no public figures for GLiDE, and "adaptive thinking" suggests some decisions will take longer.
- Control. Both are closed and available only through the vendor's API (MarkTechPost). In both cases your data goes to an outside provider.
For a small business, a model that makes simple decisions (sorting emails, routing requests, flagging odd invoices) is worth more for its price and predictability than for 7 points on a test. If you already started with Jev and it works, you have no reason to switch today.
Be careful with your data
Decision models usually receive the real content of an email or request. Before connecting one, remove customers' personal data (names, phone numbers, ID numbers, addresses) or replace it with codes. Do not send contracts, passwords or sensitive financial information. Ask the vendor where data is processed, whether it is stored, and whether it is used for training. GDPR requires you to be able to explain this, especially when the vendor is outside the European Union.
What to do on Monday
Pick one repetitive task with 3 to 5 clear options, such as sorting incoming emails. Write 50 real (anonymized) examples with the correct answer, and when you test any model, Jev or GLiDE, compare it against these examples. Your own test on your own data is worth more than any ranking made by a vendor.