Qwen3-Next-80B-A3B-Instruct Q4_K_M needs 47.29 GiB

Needs 47.29 GiB: Qwen3-Next-80B-A3B-Instruct Q4_K_M at 8,192 tokens is an estimate of 45.96 GiB–47.29 GiB. Checkpoint Qwen/Qwen3-Next-80B-A3B-Instruct. File is 48,410,988,384 bytes. No device is selected, so this is demand only, not a fit.

Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Qwen): 45.96 GiB–47.29 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.

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

Runs on

Can Qwen3-Next-80B-A3B-Instruct run on Apple M2 Max?

Publisher and provenance

PublisherQwen
Hugging FaceQwen/Qwen3-Next-80B-A3B-Instruct
Revision9c7f2fbe84465e40164a94cc16cd30b6999b0cc7
Total parameters81,324,862,720
Active parameters3,000,000,000
Layers48
Query heads16
KV heads2 (gqa)
Head dimension256
Recorded context capability262,144
Completenesscomplete · full

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: 45.09 GiB (48,410,988,384 bytes)
Raw KV cache: 0.75 GiB (805,306,368 bytes)
Engine extras / workspace: 0.13 GiB–1.45 GiB
Total estimated demand: 45.96 GiB–47.29 GiB

Published artifacts

Quantization	Selected files	Bytes	Engine	Evidence
BF16 safetensors bf16 Calculator: BF16 safetensors	41 safetensors shards Total selected bytes: 162,649,725,440 (151.48 GiB / 162.65 GB) First shard: model-00001-of-00041.safetensors Last shard: model-00041-of-00041.safetensors View all 41 shard files `model-00001-of-00041.safetensors` `model-00002-of-00041.safetensors` `model-00003-of-00041.safetensors` `model-00004-of-00041.safetensors` `model-00005-of-00041.safetensors` `model-00006-of-00041.safetensors` `model-00007-of-00041.safetensors` `model-00008-of-00041.safetensors` `model-00009-of-00041.safetensors` `model-00010-of-00041.safetensors` `model-00011-of-00041.safetensors` `model-00012-of-00041.safetensors` `model-00013-of-00041.safetensors` `model-00014-of-00041.safetensors` `model-00015-of-00041.safetensors` `model-00016-of-00041.safetensors` `model-00017-of-00041.safetensors` `model-00018-of-00041.safetensors` `model-00019-of-00041.safetensors` `model-00020-of-00041.safetensors` `model-00021-of-00041.safetensors` `model-00022-of-00041.safetensors` `model-00023-of-00041.safetensors` `model-00024-of-00041.safetensors` `model-00025-of-00041.safetensors` `model-00026-of-00041.safetensors` `model-00027-of-00041.safetensors` `model-00028-of-00041.safetensors` `model-00029-of-00041.safetensors` `model-00030-of-00041.safetensors` `model-00031-of-00041.safetensors` `model-00032-of-00041.safetensors` `model-00033-of-00041.safetensors` `model-00034-of-00041.safetensors` `model-00035-of-00041.safetensors` `model-00036-of-00041.safetensors` `model-00037-of-00041.safetensors` `model-00038-of-00041.safetensors` `model-00039-of-00041.safetensors` `model-00040-of-00041.safetensors` `model-00041-of-00041.safetensors`	162,649,725,440 (151.48 GiB / 162.65 GB)	`vllm, sglang, unspecified`	`file-metadata`
GGUF Q4_K_M (Qwen) gguf-q4_k_m Calculator: GGUF Q4_K_M (Qwen)	Qwen3-Next-80B-A3B-Instruct-Q4_K_M.gguf	48,410,988,384 (45.09 GiB / 48.41 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q5_K_M (Qwen) gguf-q5_k_m Calculator: GGUF Q5_K_M (Qwen)	Qwen3-Next-80B-A3B-Instruct-Q5_K_M.gguf	56,710,369,120 (52.82 GiB / 56.71 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q6_K (Qwen) gguf-q6_k Calculator: GGUF Q6_K (Qwen)	Qwen3-Next-80B-A3B-Instruct-Q6_K.gguf	65,528,461,152 (61.03 GiB / 65.53 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q8_0 (Qwen) gguf-q8_0 Calculator: GGUF Q8_0 (Qwen)	Qwen3-Next-80B-A3B-Instruct-Q8_0.gguf	84,812,052,320 (78.99 GiB / 84.81 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
INT4 theoretical	No published file	40,662,431,360 (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	152.29 GiB–154.17 GiB	152.85 GiB–154.75 GiB	155.10 GiB–157.07 GiB
GGUF Q4_K_M (Qwen)	45.40 GiB–46.68 GiB	45.96 GiB–47.29 GiB	48.21 GiB–49.70 GiB
GGUF Q5_K_M (Qwen)	53.13 GiB–54.56 GiB	53.69 GiB–55.17 GiB	55.94 GiB–57.58 GiB
GGUF Q6_K (Qwen)	61.34 GiB–62.94 GiB	61.90 GiB–63.55 GiB	64.15 GiB–65.96 GiB
GGUF Q8_0 (Qwen)	79.30 GiB–81.26 GiB	79.86 GiB–81.87 GiB	82.11 GiB–84.28 GiB
INT4 theoretical Theoretical rows are not downloadable artifacts.	38.18 GiB–49.57 GiB	38.74 GiB–50.27 GiB	40.99 GiB–53.04 GiB

Raw KV contribution

Standard GQA/MHA raw KV is 2 × layers × kvHeads × headDim × tokens × sequences × bytes per element. This page uses KV heads (2), not query heads (16). Using query heads would overstate this cache by 8×.

8,192 tokens, 1 sequence, FP16: 805,306,368 bytes (0.7500 GiB)
32,768 tokens: 3,221,225,472 bytes (3.00 GiB)

Open the calculator with this model

Sources

Qwen3-Next-80B-A3B-Instruct model card
config.json — revision 9c7f2fbe84465e40164a94cc16cd30b6999b0cc7
safetensors index — metadata.total_size=162649725440
Artifact repo Qwen/Qwen3-Next-80B-A3B-Instruct-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.
This MoE lists 81324862720 total parameters and 3000000000 active parameters. Active-parameter compute is not used as the weight-residency figure.
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-Next-80B-A3B-Instruct-Q4_K_M.gguf.
Multi-token prediction / draft weights exist and are not included unless enabled.
Context budget is the number of cached tokens per sequence (prompt plus reserved generation).

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.