The Sequence Learning Loop - Issue 925: Learn About Fable and Mythos 5.1, GLM-5.3-Flash, and Qwen 3.8
TheSequence Jesus Rodriguez
Anthropic, Zhipu, and Alibaba all shipped new models last month. They’re all chasing the same thing: models that can work for hours without costing a fortune.
Based on reporting by TheSequence, Jesus Rodriguez — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The past month brought three releases that matter less for leaderboard bragging rights than for what they say about where AI is headed. Anthropic, Zhipu, and Alibaba each pushed on the same problem from a different angle: how to build a model that can keep going for hours, and how to do it without turning every run into a budget fire drill.
Anthropic shipped Claude Fable 5.1 and Mythos 5.1, which are the same weights sold under two different safeguard regimes. That split is the point. The company is trying to offer one model while changing how tightly it is wrapped, which says plenty about how safety, product packaging, and customer needs are getting knotted together.
Zhipu went a different route with GLM-5.3-Flash. It is a 320B model, but only 18B parameters activate at a time, and it spent a week serving anonymous traffic on Chinese chips. That makes it less of a raw-size story than a systems story: the architecture and the hardware matter just as much as the model count on the tin.
Alibaba’s Qwen 3.8 family adds another twist. It includes the first Max-class Qwen with open weights, plus a preview of the Qwen 4 architecture. So while Anthropic is slicing access by safeguards and Zhipu is showing off efficiency on domestic hardware, Alibaba is pushing openness at the top end of the family and teasing the next version at the same time.
Taken together, these releases point in the same direction. The race is no longer just about who can make the biggest model. It is about who can keep a model running, on what chips, under what rules, and at what cost. That’s a very different kind of competition, and a much harder one to fake.
My take — AI-written commentary, not fact-checked reporting
The real story is not the shiny model names; it’s that everyone is now selling control as much as capability. Closed labs call it safety, open camps call it access, and the customer still gets the bill either way. The boring part—hours-long work, chip constraints, and packaging—has become the whole game, which is probably the first honest thing this industry has done in years.
Read more about this at: TheSequence
