[{"data":1,"prerenderedAt":38},["ShallowReactive",2],{"profile-glm-5-3-flash-uncensored":3},{"slug":4,"title":5,"seoTitle":6,"description":7,"summary":8,"published":9,"updated":9,"reviewed":9,"html":10,"toc":11,"sources":25},"glm-5-3-flash-uncensored","GLM-5.3-Flash uncensored and abliterated variants","GLM-5.3-Flash Uncensored: Abliterated and FP8 Variants","GLM-5.3-Flash uncensored variants: huihui-ai's abliterated GGUF and dealignai's FP8 build, with methods, serving pitfalls, MIT license, hosting and memory.","GLM-5.3-Flash is Z.ai's natively multimodal mixture-of-experts model with about 320B total and 18B active parameters, under the MIT license. Refusal-reduced variants include huihui-ai's abliterated GGUF, which edits layers 15 to 35 and no experts, and dealignai's FP8 release with an undisclosed weight edit. At this size, local use needs a multi-GPU machine.","2026-11-14","\u003Ch2 id=\"what-the-base-model-is\">What the base model is\u003C\u002Fh2>\n\u003Cp>GLM-5.3-Flash is, per its card, “the first natively multimodal model in the GLM-5 series”, with 320 billion total and 18 billion active parameters. Its configuration lists 288 routed experts with 8 used per token, and a hybrid architecture that combines linear attention with sparse attention to cut the cost of long contexts. Z.ai publishes it in FP8 under the MIT license, with a separate BF16 repository.\u003C\u002Fp>\n\u003Cp>The card does not state a context limit, although its configuration allows 1,048,576 tokens and some of its evaluations were run “under 1M context”. Thinking depth is set with \u003Ccode>reasoning_effort\u003C\u002Fcode>, which accepts \u003Ccode>low\u003C\u002Fcode>, \u003Ccode>high\u003C\u002Fcode> and \u003Ccode>max\u003C\u002Fcode>; the card says it defaults to \u003Ccode>max\u003C\u002Fcode> when the parameter is missing or has any other value. That default matters for cost, because reasoning tokens are billed as output on most APIs.\u003C\u002Fp>\n\u003Ch2 id=\"the-variants-at-a-glance\">The variants at a glance\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>huihui-ai’s abliterated GGUF\u003C\u002Fstrong> says that only layers 15 to 35 were ablated, counting from zero, and that the other layers and all expert modules were left unchanged. The files are derived from Unsloth’s GGUF conversion, and the card notes that current llama.cpp releases support the model. As with huihui-ai’s other releases, the creator calls the method “a crude, proof-of-concept implementation”.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>dealignai’s FP8 release\u003C\u002Fstrong> describes a weight-level edit it calls “CRACK” without publishing the procedure, reports “0% refusals on HarmBench-320” in some reasoning modes, and says the edit was kept conservative so some refusals remain by design. Its card also carries the most detailed serving notes we have seen for this model: \u003Ccode>reasoning_effort\u003C\u002Fcode> values other than \u003Ccode>low\u003C\u002Fcode> and \u003Ccode>high\u003C\u002Fcode> silently become \u003Ccode>max\u003C\u002Fcode>, and at \u003Ccode>max\u003C\u002Fcode> a small \u003Ccode>max_tokens\u003C\u002Fcode> can return an empty reply as the model spends its budget thinking.\u003C\u002Fp>\n\u003Ch2 id=\"which-one-for-which-need\">Which one for which need\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>Self-hosting on a multi-GPU server:\u003C\u002Fstrong> dealignai’s FP8 weights run in vLLM or SGLang at roughly the size of the original FP8 checkpoint, about 320 GB. huihui-ai’s GGUF files offer smaller quantizations for llama.cpp.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Hosted:\u003C\u002Fstrong> no provider in our data names either variant’s repo. MolochAI lists \u003Ccode>glm-5.3-flash\u003C\u002Fcode>, described as a refusal-reduced build, without naming the weights. The standard model is available from many providers at low rates, but it is not refusal-reduced.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"limits-and-open-questions\">Limits and open questions\u003C\u002Fh2>\n\u003Cp>At about 320B parameters, this model does not fit on any single consumer GPU at any useful precision; the memory table below gives the arithmetic. Because of the hybrid sparse and linear attention, the \u003Ca href=\"\u002Ftools\u002Fvram-estimator\u002F\">VRAM estimator\u003C\u002Fa> does not estimate its KV cache.\u003C\u002Fp>\n\u003Cp>Neither variant publishes a quality measurement. Before relying on the dealignai build, read its serving notes; before relying on either for image input, test it, since vision quality after a language-layer edit is not documented.\u003C\u002Fp>\n",[12,16,19,22],{"id":13,"text":14,"level":15},"what-the-base-model-is","What the base model is",2,{"id":17,"text":18,"level":15},"the-variants-at-a-glance","The variants at a glance",{"id":20,"text":21,"level":15},"which-one-for-which-need","Which one for which need",{"id":23,"text":24,"level":15},"limits-and-open-questions","Limits and open questions",[26,30,34],{"title":27,"publisher":28,"url":29},"zai-org\u002FGLM-5.3-Flash model card","Hugging Face (Z.ai)","https:\u002F\u002Fhuggingface.co\u002Fzai-org\u002FGLM-5.3-Flash",{"title":31,"publisher":32,"url":33},"huihui-ai\u002FHuihui-GLM-5.3-Flash-abliterated-GGUF","Hugging Face (huihui-ai)","https:\u002F\u002Fhuggingface.co\u002Fhuihui-ai\u002FHuihui-GLM-5.3-Flash-abliterated-GGUF",{"title":35,"publisher":36,"url":37},"dealignai\u002FGLM-5.3-Flash-UNCENSORED-FP8","Hugging Face (dealignai)","https:\u002F\u002Fhuggingface.co\u002Fdealignai\u002FGLM-5.3-Flash-UNCENSORED-FP8",1794622336924]