Generalized linear model

4.

GLM 5.3 The Flash, aka 0xAlpha, by all reports a solid step forward for Z.AI. 5.

OpenAI's Astra, reported to be coming as early as today.

Coverage of Mythos 5.1 and Fable 5.1 begins tomorrow.

By default I will deal with Fable 5.1 first, then Ustra.

I have decided to preliminarily offer only minimal coverage for Gemini 3.8 Flash, Mew Spark 1.3 and GLM 5.3 Air Flash unless we see more talk about them.

If they were game changers, we would see signs of that.

By all means try them to see if they make sense for you, if the match seems promising, but none of them look to be moments, or to upend the game board.

So I've got sub-bots that have other capabilities.

So right now it's installing, like, the new, um, GLM model, another one's installing and testing an Alibaba model.

One of them is optimizing, um, the Alibaba 27B model to run faster on a 5090.

So it added 76% to performance at 64K context.

Uh, I think the Gemini models, the way they are used now is for organizing data.

Like, you have to track any type of modality of data, and Flash is a perfectly decent model.

But it's not as good as the Chinese models, especially the new GLM Flash that has just come out that's 10 times cheaper for the same performance.

Um, Google did do a preview of Gemini 3.5 Pro, but it just couldn't keep up.

it was like, "Dudes, it isn't Gemini.

Gemini, it isn't." Um, people had all sorts of ideas, but more and more they converged on ZAI, whose model, uh, 5.2, GLM 5.2, was very good.

They then released 5.3, which was, they said the same model, the same LLM, but post-training was improved, and it was.

It was very good.

Uh, it runs, uh, in, uh, FP- f- actually it ran in FP8 on the Sparks, which is the, uh, higher resolution version.

But I, uh, then downgraded it to FP4 because it couldn't finish the h- co- the hof- tough coding ones.

And so I still haven't benchmarked FP4 because ZAI said, "Oh yeah, OxAlpha, that's, uh, GLM 5.3 flash."

Oh.