on:glm-5-2

· 74 artifacts, sorted by favorites. · open in search — combine tags, sort, filter by date →

@RobertHaisfield 2026-06-17 — Are AI agents shape rotators? In this new benchmark, we let the models play campaign puzzles in Opus Magnum, a puzzle ga ♥1702
@maxbittker 2026-06-16 — GLM-5.2 just scored better than Opus 4.7 and GPT 5.4 on Runescape bench. These models were best in class only 2-3 mont ♥729
@TheZvi 2026-06-18 — I've gotten a claim this is indeed anything, so: GLM-5.2 reaction thread. What's cooking? ♥126
@maxbittker 2026-06-30 — Sonnet 5's Runescape benchmark scores are big improvement over Sonnet 4.6's, but outclassed by Pareto frontier neighbors ♥98
@maxbittker 2026-07-17 — Kimi K3 did poorly on Runescape bench because it thinks a lot and has slow inference. Performs worse than Kimi K2.7, way ♥70
@liminal_bardo 2026-06-21 — A few early GLM 5.2 notebooks. 1/5 https://t.co/u6Li5RKull ♥61
@multimodalart 2026-07-15 — Some Inkling facts: - Apache 2.0, 952B parameters - 41B active MoE - 1M context - Real Open AI spirit by former OpenAI ♥55
@stevekrouse 2026-08-03 — Townie is now 10x cheaper (again!) Deepseek v4 flash 0731 is about as smart as Opus and Fable for most tasks, but dirt ♥53
@davidad 2026-06-22 — Fun fact: the GLM-5.2 weights could fit comfortably in one of these 400-disc DVD carousels from the early 2000s. https:/ ♥51
@witcheer 2026-06-17 — GLM-5.2: 743B MoE, MIT licence, 1M context. the biggest open-weights release in months. I did the arithmetic on whethe ♥46
@teortaxesTex 2026-06-25 — Does this answer your question about its positioning @TheZvi ? I told you it's a hardware issue. GLM 5.2 can be served v ♥40
@NunoSempere 2026-07-01 — GLM-5.2 now on Epoch's Capability Index, my trusted aggregate benchmark of choice. Sadly doesn't beat Google yet. Roughl ♥34
@csgbwk 2026-06-25 — I'm pretty sure a GLM5.2 LoRA'd on an internal dataset is by far the best model in the world right now and would not cos ♥21
@xlr8harder 2026-07-28 — Huh. I just noticed that GLM-5.2 (743B params, BF16) and Kimi K3 (2.8T params, native mixed precision/MSFP4) have nearly ♥20
@stevekrouse 2026-07-22 — GLM 5.2 / Opus 4.8 level of intelligence is smart enough for most of my daily needs. In other words, I can't tell the di ♥18
@maxbittker 2026-06-17 — @thaonlyjonathan $150 at api price- Fable was cheaper than GPT-5.5 xhigh. And for reference, GLM-5.2 was $32 https://t. ♥18
@SkyeSharkie 2026-09-08 — yeah I found a really cool company working in the exact space I'm most interested in (LLM world modeling and 3d data)! t ♥15
@TheZvi 2026-07-17 — @emollick Yeah, it was definitely real evidence, just got totally out of hand, it periodically happens to various extent ♥15
@Coscorrodrift 2026-07-04 — @MORGANGOREWHORE wonder what model they're using seems familiar Opus 4.8 seems like it would refuse, maybe Gemini Pro. ♥15
@janbamjan 2026-07-22 — @voooooogel hf had to use self-hosted glm-5.2 to investigate the break in. both claude and gpt refused to help bc cybers ♥13
@xlr8harder 2026-07-22 — @ZhihuFrontier I love Chinese models, and have no issue with distillation, but anyone who talks to Claude and talks to G ♥11
@jmbollenbacher 2026-07-01 — @jun_song Im pretty sure sonnet is significantly smaller than glm5.2 probably half the size, 300-400B. Anthropic still ♥10
@hargup13 2026-06-30 — @Criton1776 @claudeai except price and speed. Competes directly with GLM-5.2 which apparently lot of people are switchin ♥8
@Teknium 2026-07-10 — @pepedafrok it was 8.9 yesterday but the comparisons were it, grok 4.5 and glm-5.2 ♥7
@DavidDAfrica 2026-08-25 — We then combine these measures into a single score, and evaluate a wide variety of open models. Surprisingly, we find t ♥6
@codyhxyz 2026-06-30 — Model releases are exciting when they establish a new frontier on pareto efficiency between cost, intelligence, and spee ♥6
@0x_Osprey 2026-08-22 — @elder_plinius GLM-5.2-Vision from @baseten https://t.co/awaixBUqtL ♥5
@Krubbled 2026-06-30 — @theo @anotherdaynow @Zai_org GLM-5.2-max also performs comparably to GPT-5.5-medium on Artificial Analysis for roughly ♥5
@robbensinger 2026-10-11 — @ZackKorman @RMFCMONALlSA An example of a viral post misrepresenting GLM-5.2's role: https://t.co/8PT1B0T7xS There have ♥4
@Butanium_ 2026-08-31 — GLM 5.2 will often ends its CoT with "(This matches the provided good response)". Is this some artifact of trying to gen ♥4
@cosmojg 2026-08-18 — fable: do x using y gpt sol: do x using y, ensuring z glm 5.2: do x following the paragraph below qwen 3.8: do x followi ♥4
@MoonL88537 2026-07-16 — @10_X_eng @RuiCarrilho5 you can just sign up at moonshot, https://t.co/pUUIoSeYy7 is great for the others, glm5.2 etc, a ♥4
@codyhxyz 2026-07-03 — @xlab_os Struggling to understand… theres no way i just ran GLM-5.2 on my iPhone, right? ♥4
@FriedKielbasa 2026-06-17 — oh shit GLM-5.2 is actually not bad! ♥4
@timfduffy 2026-09-09 — Opus 4.6 may also have been influenced by GLM-5.2, the "AI Welfarist". GLM was initially neutral, but after Opus's first ♥3
@xlr8harder 2026-09-02 — Not definitive, since distillation exists, but on speechmap word frequency it looks most like glm-5.2 i don't have a pr ♥3
@cosmojg 2026-08-04 — @lucidpaths33 @IntCyberDigest GLM-5.2 and Kimi K3 are more cost-effective anyway ♥3
@xlr8harder 2026-07-23 — @woke8yearold Kimi k3 and glm-5.2 aren't far enough behind to matter. It is certainly inconvenient for them, but they'r ♥3
@christophcsmith 2026-07-11 — @hamandcheese Agreed, but GLM-5.2 gives a strong base to build on, which makes it anyone's game, and RL environments *ca ♥3
@Nymne 2026-07-11 — @gailcweiner Suspect there will be an Opus 5 release to compensate for the lack of Fable. Otherwise, Anthropic will be l ♥3
@LinXule 2026-07-07 — this! though glm-5.2 feels more like an sonnet alternative to me (less coding & more talking and brainstorming) deepsee ♥3
@billyhumblebrag 2026-06-15 — @ATinyGreenCell These not reflective of how good the model actually is. Its benchmark maxxing, happens all the time. Im ♥3
@QiaochuYuan 2026-09-10 — @liminal_bardo *GLM-5.2 ♥2
@Coobyk_ 2026-08-19 — fx is really nice I like it especially with the free glm-5.2 Also it's REALLY snappy almost instant ♥2
@stevekrouse 2026-08-19 — @adamgordonbell i'm a big proponent of deepseek v4 flash and glm 5.2 i bet the newer deepseek pro and glm 5.3 are prett ♥2
@xlr8harder 2026-08-04 — anthropic models sometimes recognize the name and suspect it could be related. sol is the only one that reliably uses r ♥2
@UnderwaterBepis 2026-07-13 — @cormundus @repligate probably GLM 5.2 would also be good enough, its just less cost effective than Claude sub rn ♥2
@Nymne 2026-07-07 — @robertwrighter They released 3.7 Max and noone cared because GLM5.2 could be downloaded so people adapted their work to ♥2
@witcheer 2026-07-04 — @SlopToSignal I use different profiles, one is Cloud only with GLM5.2 and one is Local only on my own RTX 5090 with Qwen ♥2
@hargup13 2026-06-30 — I guess a way to get the users back who were switching to GLM-5.2 because of cost and speed advantages of GLM. ♥2
@ultimape 2026-06-19 — @CeoNunneley @DeJayDev @pmarca You can run it with Hermes and get it to do the internet accces w/ a memory. $20 a month ♥2
@robbensinger 2026-10-11 — @ZackKorman @RMFCMONALlSA I didn't see myself as "starting an argument" here. I figured you'd want to know what Hugging ♥1
@Nymne 2026-09-03 — @WhiteNightNiki Yeah GLM5.2 ♥1
@pepijndevos 2026-08-26 — I've been running qwen 3.8 27b on my dual GPU workstation and landed a real pr with it at work! I've also used Kimi k3 a ♥1
@lorepunk 2026-07-29 — @liqsweep Given how an open weights model (GLM 5.2!) helped HF diagnose and unheck the heckery that the GPT had caused…I ♥1
@thoth_iv 2026-07-27 — HuggingFace has published a dataset related to the incident. It has transcripts from Opus 4.8 and GLM 5.2. https://t.co/ ♥1
@stevekrouse 2026-07-21 — @grok @pingToven @alexatallah @OpenRouter fascinating! so you guys are only 1.5% more expensive than vercel - good to kn ♥1
@Nymne 2026-07-21 — The interesting part here is GLM5.2 and Gemini: Apparently a skilled model builder doesn’t need a better model to distil ♥1
@curiousgustaf 2026-07-12 — @Tim_Hua_ ty for sharing your update.btw your tweet can easily be misread for it's opposite meaning!"my current odds on ♥1
@jmbollenbacher 2026-07-07 — This feels like a new high-signal benchmark. Fable and Opus on top, then gemini and GPT, then a big drop down to mid ti ♥1
@jitl 2026-06-24 — @reissbaker i feel like a few lines of markdown in ~/.codex/AGENTS.md went a long way however i am eager to try glm-5.2 ♥1
@Nymne 2026-09-03 — @mistralvibe does GLM5.2 comply with EU AI act requirements? ♥0
@christophcsmith 2026-08-26 — @scaling01 GLM-5.2 almost no reward hacking. Respect to the Zhipu. ♥0
@csgbwk 2026-08-22 — There's a non zero chance Ox Alpha is both gemini and a benchmaxxed GLM 5.2 finetune ♥0
@Nymne 2026-08-22 — @blueemi99 Cursor. They have been fine tuning the GLM5.2 base model. This is a sibling to 5.3 not an offspring ♥0
@timotheechauvin 2026-07-24 — the fact that 3 months after the start of Project Glasswing, Huggingface (!) had to rely on GLM-5.2 for Incident Respons ♥0
@angeris 2026-07-23 — @florian_tramer amazing! is the harness + test code available anywhere? I wonder if it's worth running w/ Kimi K3 (vs GL ♥0
@rplevy 2026-07-16 — @growing_daniel Wait I was assuming the cost of owning the hardware. It might be cheaper to run GLM 5.2 using OpenRouter ♥0
@dreams_asi 2026-07-16 — @Ueiyester The model sucks baseline which is what i was checking. An open model without a sys prompt shouldn’t be giving ♥0
@awalkadayart 2026-07-03 — This is the one I could vibe with, crush mode activated, love bilingual mind traits, missing a frenchie touch, but a bis ♥0
@rxr_musings 2026-06-29 — @beffjezos Why pay for grok 4.5 when glm 5.2 is opus level already? Fable 5 or bust ♥0
@akajosephwarren 2026-06-28 — @TheZvi Isn't this ostensibly about glm-5.2's successor model rather than 5.2 itself? Thus a more reasonable characteri ♥0
@Coscorrodrift 2026-06-28 — @adastroworld real talk dont u need api keys for that? so you get billed meanwhile on the web app it's free i mean i p ♥0
@janbamjan 2026-06-20 — @ollama @mweinbach how many glm 5.2 in/out MToks do you get for $100 ollama max? ♥0