the deepseek and glm flash on mistral's sub are definitely quants oh my god
November 11, 2026 at 5:06 AM
the deepseek and glm flash on mistral's sub are definitely quants oh my god
Call me for new GLM or DeepSeek...
November 9, 2026 at 7:08 PM
Call me for new GLM or DeepSeek...
turns out all the embargoed android security releases through may 2027 can be recovered in less than an hour by glm-5.3-flash comparing certain binaries. if anyone cares
November 9, 2026 at 12:26 PM
turns out all the embargoed android security releases through may 2027 can be recovered in less than an hour by glm-5.3-flash comparing certain binaries. if anyone cares
"GLM, make practical EUV process nodes"
Come on now.
Come on now.
But if you're a Chinese LLM company, you still follow the path of least resistance: you want to catch up in LLM capability, not have your scientists learning how to make good chips
That changes if you can't get the chips anymore
That changes if you can't get the chips anymore
November 9, 2026 at 6:38 PM
"GLM, make practical EUV process nodes"
Come on now.
Come on now.
Nex-N2.6 🔥 New multimodal agentic model from Shanghai Innovation Institute
huggingface.co/nex-agi/Nex-...
- Based on GLM-5.3-Flash
- 90% less repetitive reasoning (16% >> 1.6%)
- 100% tool list adherence
- Apache 2.0
huggingface.co/nex-agi/Nex-...
- Based on GLM-5.3-Flash
- 90% less repetitive reasoning (16% >> 1.6%)
- 100% tool list adherence
- Apache 2.0
November 12, 2026 at 1:21 PM
Nex-N2.6 🔥 New multimodal agentic model from Shanghai Innovation Institute
huggingface.co/nex-agi/Nex-...
- Based on GLM-5.3-Flash
- 90% less repetitive reasoning (16% >> 1.6%)
- 100% tool list adherence
- Apache 2.0
huggingface.co/nex-agi/Nex-...
- Based on GLM-5.3-Flash
- 90% less repetitive reasoning (16% >> 1.6%)
- 100% tool list adherence
- Apache 2.0
Small discovery from today’s debugging 🤧😂
Opus 5.5 went completely blind with those tiny micro-fixes… while GLM 5.3 Flash spotted the issue on the first run. 😭
Sometimes the cheaper/faster model just randomly saves the day. 🤝
#AI #AICoding #GLM #BuildInPublic #SoloFounder
Opus 5.5 went completely blind with those tiny micro-fixes… while GLM 5.3 Flash spotted the issue on the first run. 😭
Sometimes the cheaper/faster model just randomly saves the day. 🤝
#AI #AICoding #GLM #BuildInPublic #SoloFounder
November 12, 2026 at 2:58 PM
Small discovery from today’s debugging 🤧😂
Opus 5.5 went completely blind with those tiny micro-fixes… while GLM 5.3 Flash spotted the issue on the first run. 😭
Sometimes the cheaper/faster model just randomly saves the day. 🤝
#AI #AICoding #GLM #BuildInPublic #SoloFounder
Opus 5.5 went completely blind with those tiny micro-fixes… while GLM 5.3 Flash spotted the issue on the first run. 😭
Sometimes the cheaper/faster model just randomly saves the day. 🤝
#AI #AICoding #GLM #BuildInPublic #SoloFounder
have i shilled glm 5.3 flash enough? its still claduish but nowhere near as bad, and feels like its actually capable of doing things outside its training set a little
November 11, 2026 at 6:08 AM
have i shilled glm 5.3 flash enough? its still claduish but nowhere near as bad, and feels like its actually capable of doing things outside its training set a little
Elon Musks intelligence honeytrap is a GLM-5.3 pretending to be his grok girlfriend
November 10, 2026 at 8:52 PM
Elon Musks intelligence honeytrap is a GLM-5.3 pretending to be his grok girlfriend
What do you think the difference is, exactly? A GLM 5.3 request uses ~1.22 Wh at Neuralwatt. An RTX 5060 maxes out at 145W. So playing a graphically-intensive video game for an hour is roughly 100x worse than asking an LLM a question
November 11, 2026 at 12:32 AM
What do you think the difference is, exactly? A GLM 5.3 request uses ~1.22 Wh at Neuralwatt. An RTX 5060 maxes out at 145W. So playing a graphically-intensive video game for an hour is roughly 100x worse than asking an LLM a question
Why is GLM 5.3 not a vision model, but GLM 5.3 Flash one? Why not having vision in your best model?
November 11, 2026 at 12:23 PM
Why is GLM 5.3 not a vision model, but GLM 5.3 Flash one? Why not having vision in your best model?
Question: these models you can run locally, how are they trained? Do they in any respect, use data scraped illegally from the internet?
Cause a quick search about GLM tells me there was controversy surrounding it ripping off Claude by using its reasoning data and outputs.
They seem just as shit.
Cause a quick search about GLM tells me there was controversy surrounding it ripping off Claude by using its reasoning data and outputs.
They seem just as shit.
November 10, 2026 at 11:34 AM
Question: these models you can run locally, how are they trained? Do they in any respect, use data scraped illegally from the internet?
Cause a quick search about GLM tells me there was controversy surrounding it ripping off Claude by using its reasoning data and outputs.
They seem just as shit.
Cause a quick search about GLM tells me there was controversy surrounding it ripping off Claude by using its reasoning data and outputs.
They seem just as shit.
When a GLM 5.3 Flash run goes sideways, switch to GLM 5.3 instead of re-prompting. This guide covers which GLM model fits which coding job.
https://imagera.ai/blog/glm-coding?utm_source=bluesky&utm_medium=social&utm_campaign=text_post&utm_content=glm-coding
https://imagera.ai/blog/glm-coding?utm_source=bluesky&utm_medium=social&utm_campaign=text_post&utm_content=glm-coding
November 8, 2026 at 9:07 PM
When a GLM 5.3 Flash run goes sideways, switch to GLM 5.3 instead of re-prompting. This guide covers which GLM model fits which coding job.
https://imagera.ai/blog/glm-coding?utm_source=bluesky&utm_medium=social&utm_campaign=text_post&utm_content=glm-coding
https://imagera.ai/blog/glm-coding?utm_source=bluesky&utm_medium=social&utm_campaign=text_post&utm_content=glm-coding
to celebrate exactly 14 days since project formation: token counts
billions must decompile
billions must decompile
November 12, 2026 at 1:08 AM
to celebrate exactly 14 days since project formation: token counts
billions must decompile
billions must decompile
When Jensen said he loves Chinese models, I thought it was about DeepSeek and Kimi and GLM, etc……
November 9, 2026 at 10:12 AM
When Jensen said he loves Chinese models, I thought it was about DeepSeek and Kimi and GLM, etc……
Zhipu’s GLM 5.3 (753B MoE) for coding + long-horizon agent work is landing on Amazon Bedrock, with prompt caching and cross-Region inference. 🧠⚡️ #AI #Bedrock #OpenWeight #Coding #Agentic
Introducing GLM 5.3 on Amazon Bedrock | Amazon Web Services
GLM 5.3 from Z.ai is now available on Amazon Bedrock: a 753B-parameter mixture-of-experts model built for coding and long-horizon agentic tasks. Learn how to invoke it with the OpenAI-compatible APIs, cut cost and latency with prompt caching, and run an authorized security test with the open-source Strix agent.
aws.amazon.com
November 12, 2026 at 10:39 AM
Zhipu’s GLM 5.3 (753B MoE) for coding + long-horizon agent work is landing on Amazon Bedrock, with prompt caching and cross-Region inference. 🧠⚡️ #AI #Bedrock #OpenWeight #Coding #Agentic
When Jensen said he loves Chinese models, I thought it was about DeepSeek and Kimi and GLM, etc……
November 9, 2026 at 10:12 AM
When Jensen said he loves Chinese models, I thought it was about DeepSeek and Kimi and GLM, etc……
... or you can just run a GLM or Qwen model on your own PC without relying without any cloud service from any of those providers at all
I know western big tech wants people to believe that they have full monopoly of this technology, but they don't
I know western big tech wants people to believe that they have full monopoly of this technology, but they don't
November 10, 2026 at 8:37 AM
... or you can just run a GLM or Qwen model on your own PC without relying without any cloud service from any of those providers at all
I know western big tech wants people to believe that they have full monopoly of this technology, but they don't
I know western big tech wants people to believe that they have full monopoly of this technology, but they don't
GLM is sm0rt but deepseek is still 10x cheaper.
November 10, 2026 at 4:02 PM
GLM is sm0rt but deepseek is still 10x cheaper.
mistral’s glm flash is such a weird quant that it destroys tool calling capability and i feel bad for finding it how amusingly it affects niri
November 9, 2026 at 9:15 PM
mistral’s glm flash is such a weird quant that it destroys tool calling capability and i feel bad for finding it how amusingly it affects niri
I think the simple fact that GLM 5.3 is opensource is a win for humanity tbh
November 8, 2026 at 4:10 AM
I think the simple fact that GLM 5.3 is opensource is a win for humanity tbh
hmmm should the person with codex daybreak blue and anthropic cvp access do the security patches? naw let’s just give some random glm and call it a day i guess
November 9, 2026 at 1:58 PM
hmmm should the person with codex daybreak blue and anthropic cvp access do the security patches? naw let’s just give some random glm and call it a day i guess
GLM-5.3 is available on SambaCloud at over 200 tokens/s under the model string GLM-5.3.
November 12, 2026 at 7:45 AM
GLM-5.3 is available on SambaCloud at over 200 tokens/s under the model string GLM-5.3.
One Claude fixed fingerprint sensor earlier and Mistral Le Chonk, GLM 5.3 and Deepseek teamed up on [redacted]
November 8, 2026 at 5:52 PM
One Claude fixed fingerprint sensor earlier and Mistral Le Chonk, GLM 5.3 and Deepseek teamed up on [redacted]
No, just paying the Mistral subscription and using the app with camera/vision when needed. Both GLM 5.3 and Large 4 in the same app (GLM now by default). These models do photo and screenshot analysis pretty good.
November 8, 2026 at 6:43 PM
No, just paying the Mistral subscription and using the app with camera/vision when needed. Both GLM 5.3 and Large 4 in the same app (GLM now by default). These models do photo and screenshot analysis pretty good.