GLM 5.5 is not released yet — let me say that up front, because as of 8 August 2026 most of what you'll read about it is guesswork dressed up as news. There are no official weights, no spec sheet, nothing on Hugging Face and nothing in Ollama's library. What we do have is a set of leaks from mid-July that I covered the day they dropped, a current GLM line from Z.ai that's moving stupidly fast, and a simple way to be running the new model within minutes of it actually shipping. This page keeps all of that honest — what's confirmed, what's leak territory, and how to be ready — and I'll update it the moment the status changes.
GLM 5.5 release status: what's confirmed, what isn't
I keep two lists for every unreleased model, and I'd suggest you do the same. Here's the confirmed list, checked at publication on 8 August 2026:
- GLM is the model family from Z.ai (Zhipu AI), and GLM-5.2 is the current open flagship.
- Leaks about a 5.5 release began circulating in mid-July 2026.
- It's not on Hugging Face and it's not in Ollama's library. I checked both today.
- Z.ai has kept shipping in the meantime — I covered GLM 5.3 in early August.
And here's the other list: parameter counts, benchmark scores, release dates, new features. Every one of those currently sits in leak territory. Until Z.ai publishes weights or an official announcement, nothing in that second list is a fact — and I'm not going to pretend otherwise just to rank for a keyword.
The leaks: covered the day they dropped
The GLM-5.5 leaks started doing the rounds in mid-July 2026, and I covered them the same day in my video "New GLM-5.5 LEAKS!" — it's embedded on this page, and it's where I'd send you for the detail. I'm deliberately not restating the claimed specs here as if they were confirmed, because that's exactly how leak numbers end up quoted as fact across forty other blogs.
My rule with leaks is simple: treat them as a signal about direction, not a spec sheet. Some leaks in this space have turned out accurate. Plenty haven't. Watch the video, enjoy the speculation, then hold it loosely until the weights land somewhere you can actually download them.
📺 Watch: China's GLM 5.3 is INSANE! 🤯
The GLM line today: what you can actually use
While we wait, the line itself is in the strongest shape it's ever been. GLM-5.2 is Z.ai's open flagship — a coding model with a 1-million-token context window and open weights on Hugging Face, where the zai-org GLM-5.2 repo sits among the most-downloaded models on the platform, with a big ecosystem of community quantisations behind it. It's on Ollama too, including a glm-5.2 library entry, plus a cloud tag route if you're on Ollama's cloud subscription.
I didn't just read the model card. I wired GLM-5.2 into the Hermes agent through a custom provider and ran real multi-step agent work through it — my full GLM-5.2 review walks through the whole thing. The verdict was a clear yes: it can drive a full agent loop with tools, memory and skills. The 1M-token context was the headline reason — it holds an entire codebase or a full Obsidian vault at once.
Then Z.ai shipped again. I covered GLM 5.3 in early August ("China's GLM 5.3 is INSANE!"), which means the line has gone 5.1 to 5.2 to 5.3 within months. That cadence is the real story here — and it's why I take these leaks seriously rather than filing them under vapourware.
If you want to be running new models the day they drop instead of reading about them a fortnight later, the AI Profit Boardroom ships the Agent OS where a new model is a one-click profile swap. → Get day-one ready
What GLM 5.5 has to beat
I'm not going to predict what GLM 5.5 will do. What I can tell you, from months of hands-on testing, is the bar it has to clear — because its own stablemate set it. These are the questions I'll be asking the day it lands:
- Does it keep the 1M-token context? That window is the reason GLM-5.2 earned a place in my stack for whole-codebase work. Anything smaller would feel like a step backwards.
- Does it stay open-weights? The download numbers on Hugging Face tell you how much of GLM's momentum comes from people running it themselves.
- Does it run an agent loop better? GLM-5.2 already drives Hermes end to end — tools, memory, skills. The interesting question is whether the next release does that faster, cheaper or more reliably, not whether it can write a nicer poem.
- How does it stack up against the field? I've already run Longcat 2.0 against GLM-5.2, and the moment the new model ships it goes straight into my Goldie Bench side-by-sides against the current frontier, same as every release. That's where my GLM-5.2 agent-loop verdict came from in the first place — real runs on my own tasks, not screenshots of someone else's benchmarks.
📺 Watch: New GLM-5.5 LEAKS!
How to be ready on day one
Here's the practical part. Hermes — the self-improving agent framework from Nous Research that lives in your terminal, your messaging apps and your IDE — is provider-agnostic by design. It has a skill loop that saves solved problems as reusable skills, persistent memory, and no loyalty to any one lab. You add a provider profile, point it at an endpoint, and run. That architecture is exactly why a new GLM release can be swapped in the day it drops.
I documented two wiring paths when I set up GLM-5.2, and both are the obvious candidates for the next release:
- The local route: run the model through the Ollama daemon and point Hermes at it.
- The direct route: grab an API key from the Z.ai platform and add it as a custom provider.
Inside the Agent OS this gets even simpler: a new model is a profile swap, and your agent keeps its memory, skills and workflows intact while the brain underneath changes. Set that up now, while there's nothing to rush for, and day one becomes a two-minute job.
📺 Watch: GLM 5.2 + NoteBookLM is INSANE! 🤯
What to do today while you wait
Don't sit on your hands waiting for a release date nobody has. The best preparation for GLM 5.5 is getting comfortable with the current model, because the wiring will almost certainly carry over:
- Run GLM-5.2 now. My five-minute setup guide gets it inside Hermes today, and the full review covers tools, pricing and how it behaves on real agent work.
- Prefer local? My local model setup guide covers running Hermes against models on your own machine.
- Want to spend nothing? There are free API routes for Hermes that let you build the whole workflow without a card on file.
Where GLM 5.5 will appear when it ships
Three places, based on where every recent GLM release has landed. Watch these and you'll know before the news cycle does:
- Hugging Face — the zai-org organisation is where the open weights go.
- Ollama's library — where the runnable entries and tags show up.
- The Z.ai platform — for API access without hosting anything yourself.
I check these as part of my normal coverage, and this page gets updated when the status changes. Here's where everything stands right now:
| Thing | Status as of 8 August 2026 |
|---|---|
| GLM 5.5 leaks | Covered — watch the video on this page |
| Weights on Hugging Face | Not yet |
| Entry in Ollama's library | Not yet |
| Availability on the Z.ai platform | Not yet |
| My hands-on testing | Awaiting release — GLM-5.2 verdict live now |
Frequently asked questions
Is GLM 5.5 out yet?
No. As of 8 August 2026 it has not been released — no weights on Hugging Face, no entry in Ollama's library, no availability on the Z.ai platform. This page gets updated when that changes.
What do the leaks actually say?
Watch my video "New GLM-5.5 LEAKS!" on this page — I covered the claims the day they surfaced in mid-July 2026. I'm keeping the detail in the video rather than printing it here, because every one of those claims is unverified until Z.ai actually ships.
Will GLM 5.5 work with Hermes?
I can't promise anything about an unreleased model, but the odds are strongly in your favour. Hermes is provider-agnostic, and I've already wired GLM-5.2 in through both Ollama and a Z.ai API key. If the release follows the same pattern, it's a profile swap.
Is GLM-5.2 worth using while I wait?
Yes — and it's not a consolation prize. It runs a full agent loop with tools, memory and skills, and the 1M-token context holds an entire codebase at once. My review and setup guide are linked above; you could be running it this afternoon.
Where will it appear first?
Hugging Face (the zai-org organisation), Ollama's library and the Z.ai platform — the same three places the rest of the line lives. I'd check Hugging Face first, since open weights have been the pattern for the flagship releases.
My verdict right now
Honest answer: there's nothing to bench yet, so anyone handing out star ratings is reviewing their imagination. But the trajectory is real. Z.ai has gone 5.1 to 5.2 to 5.3 in a matter of months, the current flagship already runs serious agent work on a 1M-token context, and the leaks were interesting enough that I covered them the day they broke. My advice is the boring, profitable kind: ignore the hype cycle, get GLM-5.2 doing real work in your stack today, and set your agent up so that when the weights drop, switching over takes minutes. When that day comes, I'll have it inside Hermes and into the side-by-sides the same week — and this page will say so.
If you want an agent stack that's ready for the next GLM release before Z.ai even ships it, check out the AI Profit Boardroom — inside you get the Agent OS with ready-made model profiles, my GLM setups and masterclasses, daily tutorials, weekly live coaching calls, and me answering your questions personally. → Set up your day-one stack
