Qwen3.8-Flash-Next VRAM: 131.53 GiB

Qwen3.8-Flash-Next has 179,999,981,459 total parameters in this catalog. Activated parameter count is not independently sourced. Full-resident weight estimates use totalParameters. The smallest published weight file is GGUF Q4_K_M (bartowski) (bartowski/Qwen3.8-Flash-Next-GGUF) at 119,599,030,080 bytes (111.39 GiB). At 8,192 tokens, 1 sequence, FP16 KV, engine llama.cpp, the calculator estimates 111.75 GiB–131.53 GiB (119,991,197,504–141,224,470,933.2 bytes). Catalog unit assumption: Vendor-marketed GB for discrete GPU VRAM is treated as GiB (2^30 bytes), matching physical GDDR/HBM module organization. This is an explicit unit assumption. Under that rule, 16 GB is 17,179,869,184 bytes, 24 GB is 25,769,803,776 bytes, 32 GB is 34,359,738,368 bytes, 64 GB is 68,719,476,736 bytes. That smallest file is larger than each of those budgets, and no other published file for this model is smaller, so none of 16, 24, 32, or 64 GB fits a published file. The calculator total is higher than the file. Bartowski GGUF files in this catalog: GGUF Q4_K_M (bartowski), GGUF Q5_K_M (bartowski), GGUF Q6_K (bartowski). This catalog has no calibration trace and no planned benchmark for this model. The ranges above are calculator estimates, not a measured run. The allowlisted device check is Qwen3.8-Flash-Next on NVIDIA DGX Spark (GB10, 128 GB unified). Community recipes: TensorFold v0.6.1 MLX on one DGX Spark, SGLang file-backed PLE on one DGX Spark. The estimate method is the config and byte split and the methodology.

Qwen3.8-Flash-Next Q4_K_M at 8,192 tokens needs 111.75 GiB–131.53 GiB. No device is selected, so this is demand only, not a fit. Rough estimate. BF16 file: 359,999,963,128 bytes.

The Qwen/Qwen3.8-Flash-Next checkpoint ships BF16 safetensors of 359,999,963,128 bytes (335.28 GiB / 360.00 GB) in this catalog. That download size is not peak GPU memory, and the figures below are estimates.

Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (bartowski): 111.75 GiB–131.53 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.

This result is an estimate, not a measured trace.Not verifiedFile-backed weights

How this Qwen3.8-Flash-Next estimate is calculated

Catalog config for Qwen/Qwen3.8-Flash-Next (revision de4b8e4d43b917e7706784d8bb445c9af86a3540) records 48 layers, hidden size 2,560, 24 attention heads, 2 KV heads, head_dim 256, and max_position_embeddings 262,144. The hybrid spec is 12 groups (12 gated-attention layers, 36 Gated DeltaNet layers). KV uses those attention layers only. The all-layer GQA formula is not used.

The catalog MoE block lists 512 experts and 10 experts per token. expertOffload is unspecified. Activated parameter count is not independently sourced. Full-resident weight estimates use totalParameters. Full-resident weight estimates use totalParameters (179,999,981,459). Expert count does not shrink the file.

Vision tensors are present. Encoder bytes are not a separable file group, so they are not added as a second weight total. MTP is present. MTP file bytes are not separable from the main shard map.

The no-device auto-select is GGUF Q4_K_M (bartowski), 4 files from bartowski/Qwen3.8-Flash-Next-GGUF, 119,599,030,080 bytes (111.39 GiB). At 8,192 tokens, 1 sequence, FP16 KV, llama.cpp, the calculator estimates 111.75 GiB–131.53 GiB.This catalog has no calibration and no planned benchmark for this model. The range is a calculator estimate, not a measured run.

Official BF16 safetensors in this catalog are 359,999,963,128 bytes (335.28 GiB) across 131 shards. That checkpoint is not the auto-selected file. The title figure is the high end of the auto-selected total, not the BF16 file size.

Byte split at 8,192 tokens

weight bytes119,599,030,080 bytes
attention-layer KV bytes201,326,592 bytes
Gated DeltaNet state bytes56,623,104 bytes
conv state bytes0 bytes
total111.75 GiB–131.53 GiB

Attention-layer KV at 32,768 tokens is 805,306,368 bytes. Gated DeltaNet state at 32,768 tokens is 56,623,104 bytes, the same count as at 8,192 tokens, because that state does not scale with context.

Runs on

Can Qwen3.8-Flash-Next run on DGX Spark (GB10)?

Publisher and provenance

PublisherQwen
Hugging FaceQwen/Qwen3.8-Flash-Next
Revisionde4b8e4d43b917e7706784d8bb445c9af86a3540
Total parameters179,999,981,459
Active parametersUnknown / not independently sourced
Layers48
Query heads24
KV heads2 (hybrid)
Head dimension256
Recorded context capability262,144
Completenesspartial · partial

Active parameters describe tokens' routed compute, not the full resident model weight footprint. Full-resident weight estimates use total parameters.

Memory layers in this estimate

Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.

Weight artifact: 111.39 GiB (119,599,030,080 bytes)
Raw KV cache: 0.19 GiB (201,326,592 bytes)
Engine extras / workspace: 0.18 GiB–19.95 GiB
Total estimated demand: 111.75 GiB–131.53 GiB

Published artifacts

Quantization	Selected files	Bytes	Engine	Evidence
BF16 safetensors bf16 Calculator: BF16 safetensors	131 safetensors shards Total selected bytes: 359,999,963,128 (335.28 GiB / 360.00 GB) First shard: model-00001-of-00131.safetensors Last shard: model-00131-of-00131.safetensors View all 131 shard files `model-00001-of-00131.safetensors` `model-00002-of-00131.safetensors` `model-00003-of-00131.safetensors` `model-00004-of-00131.safetensors` `model-00005-of-00131.safetensors` `model-00006-of-00131.safetensors` `model-00007-of-00131.safetensors` `model-00008-of-00131.safetensors` `model-00009-of-00131.safetensors` `model-00010-of-00131.safetensors` `model-00011-of-00131.safetensors` `model-00012-of-00131.safetensors` `model-00013-of-00131.safetensors` `model-00014-of-00131.safetensors` `model-00015-of-00131.safetensors` `model-00016-of-00131.safetensors` `model-00017-of-00131.safetensors` `model-00018-of-00131.safetensors` `model-00019-of-00131.safetensors` `model-00020-of-00131.safetensors` `model-00021-of-00131.safetensors` `model-00022-of-00131.safetensors` `model-00023-of-00131.safetensors` `model-00024-of-00131.safetensors` `model-00025-of-00131.safetensors` `model-00026-of-00131.safetensors` `model-00027-of-00131.safetensors` `model-00028-of-00131.safetensors` `model-00029-of-00131.safetensors` `model-00030-of-00131.safetensors` `model-00031-of-00131.safetensors` `model-00032-of-00131.safetensors` `model-00033-of-00131.safetensors` `model-00034-of-00131.safetensors` `model-00035-of-00131.safetensors` `model-00036-of-00131.safetensors` `model-00037-of-00131.safetensors` `model-00038-of-00131.safetensors` `model-00039-of-00131.safetensors` `model-00040-of-00131.safetensors` `model-00041-of-00131.safetensors` `model-00042-of-00131.safetensors` `model-00043-of-00131.safetensors` `model-00044-of-00131.safetensors` `model-00045-of-00131.safetensors` `model-00046-of-00131.safetensors` `model-00047-of-00131.safetensors` `model-00048-of-00131.safetensors` `model-00049-of-00131.safetensors` `model-00050-of-00131.safetensors` `model-00051-of-00131.safetensors` `model-00052-of-00131.safetensors` `model-00053-of-00131.safetensors` `model-00054-of-00131.safetensors` `model-00055-of-00131.safetensors` `model-00056-of-00131.safetensors` `model-00057-of-00131.safetensors` `model-00058-of-00131.safetensors` `model-00059-of-00131.safetensors` `model-00060-of-00131.safetensors` `model-00061-of-00131.safetensors` `model-00062-of-00131.safetensors` `model-00063-of-00131.safetensors` `model-00064-of-00131.safetensors` `model-00065-of-00131.safetensors` `model-00066-of-00131.safetensors` `model-00067-of-00131.safetensors` `model-00068-of-00131.safetensors` `model-00069-of-00131.safetensors` `model-00070-of-00131.safetensors` `model-00071-of-00131.safetensors` `model-00072-of-00131.safetensors` `model-00073-of-00131.safetensors` `model-00074-of-00131.safetensors` `model-00075-of-00131.safetensors` `model-00076-of-00131.safetensors` `model-00077-of-00131.safetensors` `model-00078-of-00131.safetensors` `model-00079-of-00131.safetensors` `model-00080-of-00131.safetensors` `model-00081-of-00131.safetensors` `model-00082-of-00131.safetensors` `model-00083-of-00131.safetensors` `model-00084-of-00131.safetensors` `model-00085-of-00131.safetensors` `model-00086-of-00131.safetensors` `model-00087-of-00131.safetensors` `model-00088-of-00131.safetensors` `model-00089-of-00131.safetensors` `model-00090-of-00131.safetensors` `model-00091-of-00131.safetensors` `model-00092-of-00131.safetensors` `model-00093-of-00131.safetensors` `model-00094-of-00131.safetensors` `model-00095-of-00131.safetensors` `model-00096-of-00131.safetensors` `model-00097-of-00131.safetensors` `model-00098-of-00131.safetensors` `model-00099-of-00131.safetensors` `model-00100-of-00131.safetensors` `model-00101-of-00131.safetensors` `model-00102-of-00131.safetensors` `model-00103-of-00131.safetensors` `model-00104-of-00131.safetensors` `model-00105-of-00131.safetensors` `model-00106-of-00131.safetensors` `model-00107-of-00131.safetensors` `model-00108-of-00131.safetensors` `model-00109-of-00131.safetensors` `model-00110-of-00131.safetensors` `model-00111-of-00131.safetensors` `model-00112-of-00131.safetensors` `model-00113-of-00131.safetensors` `model-00114-of-00131.safetensors` `model-00115-of-00131.safetensors` `model-00116-of-00131.safetensors` `model-00117-of-00131.safetensors` `model-00118-of-00131.safetensors` `model-00119-of-00131.safetensors` `model-00120-of-00131.safetensors` `model-00121-of-00131.safetensors` `model-00122-of-00131.safetensors` `model-00123-of-00131.safetensors` `model-00124-of-00131.safetensors` `model-00125-of-00131.safetensors` `model-00126-of-00131.safetensors` `model-00127-of-00131.safetensors` `model-00128-of-00131.safetensors` `model-00129-of-00131.safetensors` `model-00130-of-00131.safetensors` `model-00131-of-00131.safetensors`	359,999,963,128 (335.28 GiB / 360.00 GB)	`vllm, sglang, unspecified`	`file-metadata`
GGUF Q8_0 (Unsloth) gguf-q8_0 Calculator: GGUF Q8_0 (Unsloth)	6 safetensors shards Total selected bytes: 188,225,033,248 (175.30 GiB / 188.23 GB) First shard: Q8_0/Qwen3.8-Flash-Next-Q8_0-00001-of-00006.gguf Last shard: Q8_0/Qwen3.8-Flash-Next-Q8_0-00006-of-00006.gguf View all 6 shard files `Q8_0/Qwen3.8-Flash-Next-Q8_0-00001-of-00006.gguf` `Q8_0/Qwen3.8-Flash-Next-Q8_0-00002-of-00006.gguf` `Q8_0/Qwen3.8-Flash-Next-Q8_0-00003-of-00006.gguf` `Q8_0/Qwen3.8-Flash-Next-Q8_0-00004-of-00006.gguf` `Q8_0/Qwen3.8-Flash-Next-Q8_0-00005-of-00006.gguf` `Q8_0/Qwen3.8-Flash-Next-Q8_0-00006-of-00006.gguf`	188,225,033,248 (175.30 GiB / 188.23 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q4_K_M (bartowski) gguf-q4_k_m Calculator: GGUF Q4_K_M (bartowski)	Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00001-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00002-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00003-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00004-of-00004.gguf	119,599,030,080 (111.39 GiB / 119.60 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q5_K_M (bartowski) gguf-q5_k_m Calculator: GGUF Q5_K_M (bartowski)	Qwen3.8-Flash-Next-Q5_K_M/Qwen3.8-Flash-Next-Q5_K_M-00001-of-00004.gguf, Qwen3.8-Flash-Next-Q5_K_M/Qwen3.8-Flash-Next-Q5_K_M-00002-of-00004.gguf, Qwen3.8-Flash-Next-Q5_K_M/Qwen3.8-Flash-Next-Q5_K_M-00003-of-00004.gguf, Qwen3.8-Flash-Next-Q5_K_M/Qwen3.8-Flash-Next-Q5_K_M-00004-of-00004.gguf	134,668,639,040 (125.42 GiB / 134.67 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q6_K (bartowski) gguf-q6_k Calculator: GGUF Q6_K (bartowski)	5 safetensors shards Total selected bytes: 168,113,805,248 (156.57 GiB / 168.11 GB) First shard: Qwen3.8-Flash-Next-Q6_K/Qwen3.8-Flash-Next-Q6_K-00001-of-00005.gguf Last shard: Qwen3.8-Flash-Next-Q6_K/Qwen3.8-Flash-Next-Q6_K-00005-of-00005.gguf View all 5 shard files `Qwen3.8-Flash-Next-Q6_K/Qwen3.8-Flash-Next-Q6_K-00001-of-00005.gguf` `Qwen3.8-Flash-Next-Q6_K/Qwen3.8-Flash-Next-Q6_K-00002-of-00005.gguf` `Qwen3.8-Flash-Next-Q6_K/Qwen3.8-Flash-Next-Q6_K-00003-of-00005.gguf` `Qwen3.8-Flash-Next-Q6_K/Qwen3.8-Flash-Next-Q6_K-00004-of-00005.gguf` `Qwen3.8-Flash-Next-Q6_K/Qwen3.8-Flash-Next-Q6_K-00005-of-00005.gguf`	168,113,805,248 (156.57 GiB / 168.11 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
INT4 theoretical	No published file	89,999,990,730 (theoretical)	—	theoretical

Memory by quantization and context

Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.

Quantization	2,048 tokens	8,192 tokens	32,768 tokens
BF16 safetensors	336.00 GiB–388.56 GiB	336.14 GiB–388.73 GiB	336.70 GiB–389.39 GiB
GGUF Q8_0 (Unsloth)	175.52 GiB–206.32 GiB	175.66 GiB–206.50 GiB	176.23 GiB–207.19 GiB
GGUF Q4_K_M (bartowski)	111.61 GiB–131.35 GiB	111.75 GiB–131.53 GiB	112.31 GiB–132.22 GiB
GGUF Q5_K_M (bartowski)	125.64 GiB–147.81 GiB	125.79 GiB–147.99 GiB	126.35 GiB–148.68 GiB
GGUF Q6_K (bartowski)	156.79 GiB–184.35 GiB	156.93 GiB–184.53 GiB	157.50 GiB–185.22 GiB
INT4 theoretical Theoretical rows are not downloadable artifacts.	84.04 GiB–108.65 GiB	84.18 GiB–108.83 GiB	84.75 GiB–109.52 GiB

Raw KV contribution

Open the calculator with this model

Sources

Qwen3.8-Flash-Next model card — Hybrid MoE (36 linear + 12 full). Expert offload unspecified; active params are not resident weights. File size is the packed checkpoint.
config.json — revision de4b8e4d43b917e7706784d8bb445c9af86a3540
safetensors index — metadata.total_size=359999963128
Artifact repo unsloth/Qwen3.8-Flash-Next-GGUF
Artifact repo bartowski/Qwen3.8-Flash-Next-GGUF

Assumptions that affect this result

Requested context is within the recorded native window of 262144 tokens.
Full-resident weight estimates use total parameters, including inactive MoE experts.
Activated parameter count is unknown / not independently sourced. Active parameters describe routed compute, not the full resident model weight footprint.
Weight bytes come from the selected weight files only, not from every file in the repository.
Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
Counted files: Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00001-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00002-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00003-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00004-of-00004.gguf.
This checkpoint has a vision component. Text-only inference does not include vision encoder or projector bytes.
Multi-token prediction / draft weights exist and are not included unless enabled.

Calculator

Loading catalog snapshot…

Choose a model first, then check a device.

Model

Show historical models

Quantization

Device type

Hardware

Workload

Context length primarily affects KV cache memory.

Advanced settings

Runtime

Engine and KV cache type for this estimate.

EngineKV cache typeRecurrent / conv state dtype Include theoretical estimates (expert) Allow a substitute if the filter has no match

Workload

Overhead scenario and the prefill workspace.

Overhead scenario Include heavier prefill workspace

Topology

How the model is split across devices. Memory is not added together.

DeploymentIntra-replica splitDevice count

Replica count is an advanced deployment option. Quantity is nodes or GPUs, not replicas.

Replica groupsDevices per replica

Budget

System reserve and safety margin. An empty number keeps the previous value.

System / display reserve

GiB (2^30 bytes)

Budget already excludes system reserveExtra safety margin

Custom bits per weightDisplay unit

Custom model

Import a config or type the architecture. Nothing is scanned until you import.

Import local config.json

Parsed in the browser. The file is not uploaded.

Public Hugging Face repoRevision

Only huggingface.co, config.json, no token. Private repos are rejected.

Total parametersActive parametersLayersQuery headsKV headsHead dimension

Search or pick a catalog checkpoint to see memory demand.

Search or pick a catalog checkpoint to see memory demand.

Qwen3.8-Flash-Next VRAM: 131.53 GiB

How this Qwen3.8-Flash-Next estimate is calculated

Byte split at 8,192 tokens

Runs on

Publisher and provenance

Memory layers in this estimate

Published artifacts

Memory by quantization and context

Raw KV contribution

Related pages

Sources

Assumptions that affect this result

Calculator