Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GLM-5.3-Flash

z.ai

Post reply on HN
Print view View original

481–490 of 605 posts

Re: GLM-5.3-Flash

#481

Post by skohan »

hadlock wrote: post #457
cdurth wrote: post #426

Earlier quoted context omitted.

I can't get 3.8 to exit thinking loops. It will just think and think and think on the most trivial topics. I wanted it to port a speed test powershell script to c#. Claude opus 5 completes it under 60 seconds. I let 3.8 churn about 6 different times for 30+ minutes and it never wrote a single line of code to disk. It wrote lots of lines in thinking. unsloth/Qwen3.8-27B-GGUF UD-Q3_K_XL DSH (pi) Any tips?

set thinking to minimal and use these jinja templates: https://huggingface.co/froggeric/Qwen-Fixed-Chat-Templates we went from 62% completion to 92% using a claude code harness

3.8 doesn't have a minimal thinking mode, only low, medium and xhigh.

Re: GLM-5.3-Flash

#482

Post by rescbr »

matheusmoreira wrote: post #356
glub wrote: post #286

Earlier quoted context omitted.

When they initially revoked TAC for a bunch of users due to a "technical error", the TAC verification flow opened a Persona iframe where you select the document country first. I was able to select it there (Georgia, in my case), went through the entire flow, and then got locked out after 8 attempts. Persona itself was successful end to end, so it failed somewhere on the OAI side. Other users then started reporting th…

That's horrible... For the record, I just made a second account, got verified by Persona and still didn't get into TAC. No "we are unable to verify identities in this country" message. No mention of my country whatsoever. Persona verification was successful. What else do they want from us? Were it not for Z.ai's obnoxious terms, I would have switched to them already...

For my own stuff I simply don't care if Z.ai is learning straight from me or from my GitHub repo a few days later when I push the code.

When I use it for reverse engineering, same thing, the code is already out there, but in binary form, which is becoming trivial for these LLMs to work on.

I even bought some shares from them and got some money out of it!

Re: GLM-5.3-Flash

#483

Post by rescbr »

throw10920 wrote: post #423
matheusmoreira wrote: post #422

Earlier quoted context omitted.

That's alarming. I want to use these models to red team my own computers. How are people getting around this?

> I want to use these models to red team my own computers. Exactly what I was trying to use it for! ): I'm in the same boat - I haven't heard of a way to get around it aside from either self-hosting (GLM-5.2? good luck ) or "self-hosting" (paying bucks per hour to Vast) an abliterated model.

What is the harness that you're using?

I found that GLM-5.2 was pretty happy helping me reverse engineer/hack devices.

Maybe the system prompt you're injecting is making it refuse?

Re: GLM-5.3-Flash

#484

Post by uncleocode »

preommr wrote: post #253

So the vagueposting by googlers about Ox Alpha was just... what exactly? Like I get that they have to be careful about comms, but surely senior members of the team can clarify when something is NOT them, when everyone is gosspiing it is them.

The lack of measurement causes existence of such claims or discussion. Last Friday I built this Model fingerprint calculator and I tested between OxAlpha with all other claimed models, the only match was GLM. It generates, or measures regardless of the model-weights or its training data. No need guessing when one can measure it. I tested with Gemini family too, far different. Here is the link to my experiment https://github.com/unclecode/modelprint

Re: GLM-5.3-Flash

#485

Post by alexfortin »

Kholin wrote: post #458
bel8 wrote: post #351

If you're on opencode's go $10/mo plan and want to use GLM-5.3-flash right now on pi, you can add this to models.json until pi updates to support it: { "providers": { "opencode-go": { "models": [ { "id": "glm-5.3-flash", "name": "GLM-5.3 Flash", "api": "openai-completions", "baseUrl": "https://opencode.ai/zen/go/v1", "reasoning": true, "input": ["text", "image"], "cost": { "input": 0.15, "output": 0.5, "cacheRead": 0…

For openrouter in pi: { "providers": { "openrouter": { "models": [ { "id": "z-ai/glm-5.3-flash", "name": "Z.ai: GLM 5.3 Flash", "reasoning": true, "thinkingLevelMap": { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max" }, "input": ["text", "image"], "cost": { "input": 0.075, "output": 0.25, "cacheRead": 0.015, "cacheWrite": 0 }, "contextWindow": 1048576, "maxToken…

For using (lite) Z.ai subscription in Pi while it's not available yet:

https://forge.l3x.in/alex/pi-shared/src/branch/master/extens...

Re: GLM-5.3-Flash

#486

Post by throw10920 »

rescbr wrote: post #483
throw10920 wrote: post #423

Earlier quoted context omitted.

> I want to use these models to red team my own computers. Exactly what I was trying to use it for! ): I'm in the same boat - I haven't heard of a way to get around it aside from either self-hosting (GLM-5.2? good luck ) or "self-hosting" (paying bucks per hour to Vast) an abliterated model.

What is the harness that you're using? I found that GLM-5.2 was pretty happy helping me reverse engineer/hack devices. Maybe the system prompt you're injecting is making it refuse?

> Maybe the system prompt you're injecting is making it refuse?

No, this has nothing to do with my harness. I use one of the most popular open-source harnesses available.

> I found that GLM-5.2 was pretty happy helping me reverse engineer/hack devices.

This is a completely different category of things than what I'm getting refusals on, so I'm not sure why you're bringing it up.

Re: GLM-5.3-Flash

#487

Post by paytonjjones »

glub wrote: post #185
mrngld wrote: post #79

Chinese labs are so used to manipulating benchmarks to try to flatter inferior models that when they finally have one that's really pretty good I think the official announcement here undersells it. https://deepswe.datacurve.ai/ That's pretty solid. Smarter and cheaper than Luna xhigh, not as smart but less expensive than Luna max. Smashes deepseek v4 flash, and even worse it matches v4 pro at a tiny fraction the cost…

I don't know how anyone can actually use Luna max on ANY real workload. I've had Sol orchestrate a bunch of Luna agents, these agents were explicitly given small chunks of larger objectives and they still filled their entire context windows with just reasoning tokens, until compaction hit, and then reasoning again. I've probably wasted a good 40% of my weekly usage on Luna Max agents just thinking and not writing a s…

I've had the same observation that Luna will quickly fill up its context window with reasoning, but it surprisingly hasn't been a problem really.

It will cycle through like 3 /compacts, complete the complex goal successfully, and cost me like 1% of my weekly usage on the $20 plan.

Edit: this is me using Luna directly, not Sol as the taskmaster

Re: GLM-5.3-Flash

#489

Post by freakynit »

polski-g wrote: post #372
pphysch wrote: post #278

Earlier quoted context omitted.

OAI/Anthropic shareholder? Speed and intelligence are not "every way". Cost is essential. Hence the Pareto boundary illustrated in TFA.

It literally cannot complete tasks that Luna can do easily. It doesn't matter how cheap it is.

It actually can. I have been using it regularly for past 4 days. It is on sol-low level. I deliberately tested it on a moderately complex task. glm-5.3-flash one-shotted it correctly. Luna max couldn't achieve parity even after 3 total attempts.

It was creating java bindings for this project: https://github.com/jeffhajewski/latticedb

And here is the binding one-shotted by glm-5.3-flash: https://github.com/jeffhajewski/latticedb/pull/5

Re: GLM-5.3-Flash

#490

Post by glub »

matheusmoreira wrote: post #356
glub wrote: post #286

Earlier quoted context omitted.

When they initially revoked TAC for a bunch of users due to a "technical error", the TAC verification flow opened a Persona iframe where you select the document country first. I was able to select it there (Georgia, in my case), went through the entire flow, and then got locked out after 8 attempts. Persona itself was successful end to end, so it failed somewhere on the OAI side. Other users then started reporting th…

That's horrible... For the record, I just made a second account, got verified by Persona and still didn't get into TAC. No "we are unable to verify identities in this country" message. No mention of my country whatsoever. Persona verification was successful. What else do they want from us? Were it not for Z.ai's obnoxious terms, I would have switched to them already...

Ah yeah, new accounts get instantly blocked on TAC. Their backend has 2 failure states for `id_verification_status` - `failed` and `blocked`.

Nationality bans get `failed`, new accounts (or rather, accounts with not enough good signals) get `blocked` on first attempt.

Post reply on HN