# Qwen3.5-397B-A17B-VQ-2.6bpw by Noah Zelezny: Open Model Source: https://savrn.com/models/qwen3-5-397b-a17b-vq-2-6bpw Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve Qwen3.5-397B-A17B-VQ-2.6bpw (59.2B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 118.3 GB | 142.0 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 · [1x MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) $2.59 | | 8-bit | 59.2 GB | 71.0 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 29.6 GB | 35.5 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 7, 2026. [Qwen3.5-397B-A17B-VQ-2.6bpw on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/qwen3-5-397b-a17b-vq-2-6bpw/gpus) ## Model Card By Noah Zelezny, published under apache-2.0, revision 5ff492af70af. 105.5 GiB text weights — the balanced build. Also ships the bf16 vision tower (+0.85 GiB) and an optional MTP draft head (+5.4 GiB); full download 111.8 GiB. It lands in the same size class as the strongest community quant at this rate; the table below is the comparison. A vector-quantized build of [Qwen3.5-397B-A17B](https://huggingface.co/Qwen/Qwen3.5-397B-A17B) for Apple Silicon. Stock mlx-lm, no patches — the VQ runtime ships inside the checkpoint as model.py. It sits between the VQ-2.4bpw daily driver and the VQ-3bpw quality build, and needs the same hardware class as the latter. ### Changelog #### 2026-10-01 — vq-skipzero: dead rows dropped [Read the full model card (2,705 words)](https://savrn.com/models/qwen3-5-397b-a17b-vq-2-6bpw/card) ## Configuration Architecture Qwen3_5MoeForConditionalGeneration Context length (tokens) 262,144 Layers 60 Hidden size 4,096 Attention heads 32 Key/value heads 2 Head dimension 256 Vocabulary size 248,320 Experts 512 Experts active per token 10 Model type qwen3_5_moe ## Identity and Version Repository TheDrainFlorist/Qwen3.5-397B-A17B-VQ-2.6bpw Publisher Noah Zelezny Task Text generation Modality Text Library mlx Parameters 59.2B parameters Languages en Revision 5ff492af70af6174712977788b4fb3c2a1b2d8d5 First published 2026-08-25 Last updated 2026-10-02 ## Files and Weights 44 files, 125.9 GB in total. The weights are 29 files totalling 120.0 GB in safetensors. Weights29 files · 120.0 GB Configuration8 files · 715.3 KB Tokenizer2 files · 20.0 MB Documentation1 file · 18.5 KB Other3 files · 5.8 GB Repository1 file · 1.6 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | model-00001-of-00027.safetensors | Weights | 2.9 GB | 13336296bd63 | | model-00002-of-00027.safetensors | Weights | 2.9 GB | 3f627ffffd6c | | model-00003-of-00027.safetensors | Weights | 3.5 GB | fe3fdf452290 | | model-00004-of-00027.safetensors | Weights | 4.0 GB | 1d8f8cf6c837 | | model-00005-of-00027.safetensors | Weights | 3.7 GB | 7aa0e8b6259f | | model-00006-of-00027.safetensors | Weights | 3.9 GB | 9aeaa842415c | | model-00007-of-00027.safetensors | Weights | 4.3 GB | 13599756042f | | model-00008-of-00027.safetensors | Weights | 4.2 GB | 95019aaee0f8 | | model-00009-of-00027.safetensors | Weights | 4.3 GB | 8fea418dcece | | model-00010-of-00027.safetensors | Weights | 4.6 GB | 6c33be9fe149 | | model-00011-of-00027.safetensors | Weights | 4.5 GB | 69f9c5f0c15f | | model-00012-of-00027.safetensors | Weights | 4.6 GB | fe087e399e41 | | model-00013-of-00027.safetensors | Weights | 4.8 GB | 5cecc8238fa5 | | model-00014-of-00027.safetensors | Weights | 4.8 GB | a743a8cb5ae3 | | model-00015-of-00027.safetensors | Weights | 4.8 GB | 1c495f545d4a | | model-00016-of-00027.safetensors | Weights | 4.9 GB | 855fe5458298 | | model-00017-of-00027.safetensors | Weights | 4.8 GB | efe1be6fab36 | | model-00018-of-00027.safetensors | Weights | 4.7 GB | edfa2485d70a | | model-00019-of-00027.safetensors | Weights | 4.9 GB | 497e27b264f3 | | model-00020-of-00027.safetensors | Weights | 4.7 GB | 13dcca9068cc | | model-00021-of-00027.safetensors | Weights | 4.7 GB | d49698bd239a | | model-00022-of-00027.safetensors | Weights | 4.8 GB | a4db98df474a | | model-00023-of-00027.safetensors | Weights | 4.6 GB | b398e044c6e0 | | model-00024-of-00027.safetensors | Weights | 4.5 GB | 7ef60393b775 | | model-00025-of-00027.safetensors | Weights | 3.8 GB | 83a68dcf7925 | | model-00026-of-00027.safetensors | Weights | 3.1 GB | 2f1291d47896 | | model-00027-of-00027.safetensors | Weights | 2.0 GB | 278ee2fb3735 | | model-vision-graft.safetensors | Weights | 912.1 MB | b47e83150554 | | mtp-head-q6.safetensors | Weights | 5.8 GB | 1563d297c7bc | | config.json | Configuration | 149.7 KB | — | | generation_config.json | Configuration | 244 B | — | | model.py | Configuration | 262.3 KB | — | | model.safetensors.index.json | Configuration | 288.6 KB | — | | preprocessor_config.json | Configuration | 390 B | — | | skipzero_load.py | Configuration | 5.6 KB | — | | video_preprocessor_config.json | Configuration | 385 B | — | | vqlab_provenance.json | Configuration | 8.1 KB | — | | README.md | Documentation | 18.5 KB | — | | chat_template.jinja | Other | 7.8 KB | — | | mtp-head-q6.safetensors.fp32norms-bak | Other | 5.8 GB | a5cad76f094e | | vqlab_provenance.history.jsonl | Other | 55.8 KB | — | | .gitattributes | Repository | 1.6 KB | — | | tokenizer.json | Tokenizer | 20.0 MB | 06b9509352d2 | | tokenizer_config.json | Tokenizer | 1.2 KB | — | ## License and Download License apache-2.0 Access Open weights, no gate Download size 120.0 GB [Download from Noah Zelezny](https://huggingface.co/TheDrainFlorist/Qwen3.5-397B-A17B-VQ-2.6bpw) Released by Noah Zelezny through its official repository on Hugging Face. [Read the license](https://www.apache.org/licenses/LICENSE-2.0). ## Built From - Derived from Qwen/Qwen3.5-397B-A17B - Quantized from Qwen/Qwen3.5-397B-A17B ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 120.0 GB | | 16-bit | 118.3 GB | | 8-bit | 59.2 GB | | 4-bit | 29.6 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Questions About Qwen3.5-397B-A17B-VQ-2.6bpw ### How much GPU memory does Qwen3.5-397B-A17B-VQ-2.6bpw need? About 142 GB at 16-bit and 35.5 GB at 4-bit: the weights (59.2B parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run Qwen3.5-397B-A17B-VQ-2.6bpw on? At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use Qwen3.5-397B-A17B-VQ-2.6bpw commercially? Yes. Qwen3.5-397B-A17B-VQ-2.6bpw is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors. ### What is Qwen3.5-397B-A17B-VQ-2.6bpw's context length? 262,144 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Text generation ### [less-is-moe-gpt-oss-120b-gpqa-main-64-intdim-e-50](https://savrn.com/models/less-is-moe-gpt-oss-120b-gpqa-main-64-intdim-e-50) [XINKAI ZOU](https://savrn.com/model-publishers/jayzou3773) This checkpoint was structurally pruned with the released Less-is-MoE mean-absolute-gradient method. It removes exactly 50% of routed-expert FFN neurons using 64 calibration samples from the gpqamain configuration of Idavidrein/gpqa revision 633f5ee89ab8ad4522a9f850766b73f62147ffdd. The released loader settings are preserved: train, Question plus shuffled choices, Explanation, selectionseed=1234, BF16, and no optimizer step. The samples are full length: no tokenizer maxlength, truncation, or padding. The longest input for this tokenizer is 1,511 tokens. The source MXFP4 checkpoint was explicitly dequantized to BF16 before scoring and pruning. The source-row selection hash is… Open weights apache-2.0 59.5B parameters 131,072 tokens [View model](https://savrn.com/models/less-is-moe-gpt-oss-120b-gpqa-main-64-intdim-e-50) Model · Text generation ### [Qwen3.5-397B-A17B-VQ-2.2bpw](https://savrn.com/models/qwen3-5-397b-a17b-vq-2-2bpw) [Noah Zelezny](https://savrn.com/model-publishers/thedrainflorist) 88.7 GiB text weights — the accessibility build, the roomiest fit on a 128 GB Mac. Also ships the bf16 vision tower (+0.85 GiB) and an optional MTP draft head (+5.4 GiB); full download 95.0 GiB. (v2, mixed geometry.) v2 — updated 2026-08-22. This repository now serves a rebuilt artifact at the same size and the same bits per weight, with a different codebook geometry that measures better on both perplexity corpora. v1's numbers are kept below rather than quietly overwritten, and v1's bytes remain downloadable by pinning the previous revision: A vector-quantized build of Qwen3.5-397B-A17B running? 88.7 GiB text weights — it runs on a single 128 GB Apple Silicon machine with ≈7 GiB more… Open weights apache-2.0 62.6B parameters 262,144 tokens mlx [View model](https://savrn.com/models/qwen3-5-397b-a17b-vq-2-2bpw) Model · Text generation ### [less-is-moe-qwen3.5-122b-a10b-gpqa-main-64-intdim-g-50](https://savrn.com/models/less-is-moe-qwen3-5-122b-a10b-gpqa-main-64-intdim-g-50) [XINKAI ZOU](https://savrn.com/model-publishers/jayzou3773) This checkpoint was structurally pruned with the released Less-is-MoE mean-absolute-gradient method. It removes exactly 50% of routed-expert FFN neurons using 64 calibration samples from the gpqamain configuration of Idavidrein/gpqa revision 633f5ee89ab8ad4522a9f850766b73f62147ffdd. The released loader settings are preserved: train, Question plus shuffled choices, Explanation, selectionseed=1234, BF16, and no optimizer step. The samples are full length: no tokenizer maxlength, truncation, or padding. The longest input for this tokenizer is 1,632 tokens. The source checkpoint was loaded and pruned in BF16. The source-row selection hash is… Open weights apache-2.0 64.1B parameters 262,144 tokens [View model](https://savrn.com/models/less-is-moe-qwen3-5-122b-a10b-gpqa-main-64-intdim-g-50) Model · Text generation ### [less-is-moe-qwen3.5-122b-a10b-gpqa-main-64-intdim-l-50](https://savrn.com/models/less-is-moe-qwen3-5-122b-a10b-gpqa-main-64-intdim-l-50) [XINKAI ZOU](https://savrn.com/model-publishers/jayzou3773) This checkpoint was structurally pruned with the released Less-is-MoE mean-absolute-gradient method. It removes exactly 50% of routed-expert FFN neurons using 64 calibration samples from the gpqamain configuration of Idavidrein/gpqa revision 633f5ee89ab8ad4522a9f850766b73f62147ffdd. The released loader settings are preserved: train, Question plus shuffled choices, Explanation, selectionseed=1234, BF16, and no optimizer step. The samples are full length: no tokenizer maxlength, truncation, or padding. The longest input for this tokenizer is 1,632 tokens. The source checkpoint was loaded and pruned in BF16. The source-row selection hash is… Open weights apache-2.0 64.1B parameters 262,144 tokens [View model](https://savrn.com/models/less-is-moe-qwen3-5-122b-a10b-gpqa-main-64-intdim-l-50) Model · Text generation ### [less-is-moe-qwen3.5-122b-a10b-gpqa-main-64-intdim-e-50](https://savrn.com/models/less-is-moe-qwen3-5-122b-a10b-gpqa-main-64-intdim-e-50) [XINKAI ZOU](https://savrn.com/model-publishers/jayzou3773) This checkpoint was structurally pruned with the released Less-is-MoE mean-absolute-gradient method. It removes exactly 50% of routed-expert FFN neurons using 64 calibration samples from the gpqamain configuration of Idavidrein/gpqa revision 633f5ee89ab8ad4522a9f850766b73f62147ffdd. The released loader settings are preserved: train, Question plus shuffled choices, Explanation, selectionseed=1234, BF16, and no optimizer step. The samples are full length: no tokenizer maxlength, truncation, or padding. The longest input for this tokenizer is 1,632 tokens. The source checkpoint was loaded and pruned in BF16. The source-row selection hash is… Open weights apache-2.0 64.1B parameters 262,144 tokens [View model](https://savrn.com/models/less-is-moe-qwen3-5-122b-a10b-gpqa-main-64-intdim-e-50) Model · Text generation ### [Qwen3.5-397B-A17B-VQ-3.1bpw](https://savrn.com/models/qwen3-5-397b-a17b-vq-3-1bpw) [Noah Zelezny](https://savrn.com/model-publishers/thedrainflorist) 125.3 GiB text weights — the quality build. Also ships the bf16 vision tower (+0.85 GiB) and an optional MTP draft head (+5.4 GiB); full download 131.6 GiB. A vector-quantized build of Qwen3.5-397B-A17B for machines with memory to spend: the strongest quantization we know how to make of this model at this size, on stock mlx-lm, no patches. 12.43% of the 64-element weight groups in the Qwen3.5-397B teacher sit in output rows whose weights are about 1e-29. The vq-skipzero format drops those rows' codes and scales on disk and in memory; they output exact zeros, and the live rows are byte-identical to the previous revision (vqlab sz-check, every module). Text weights go from 141.71 to 125.31… Open weights apache-2.0 64.5B parameters 262,144 tokens mlx [View model](https://savrn.com/models/qwen3-5-397b-a17b-vq-3-1bpw) ## Noah Zelezny [All models and datasets](https://savrn.com/model-publishers/thedrainflorist) ## Versions - [5ff492af70af](https://savrn.com/models/qwen3-5-397b-a17b-vq-2-6bpw/versions/5ff492af70af) · current 2026-10-02 ## Explore More - [All text generation models](https://savrn.com/models/tasks/text-generation) - [All models under apache-2.0](https://savrn.com/models/licenses/apache-2-0) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-02. - [Hugging Face record](https://huggingface.co/TheDrainFlorist/Qwen3.5-397B-A17B-VQ-2.6bpw) - [How the hub is built](https://savrn.com/model-hub/methodology)