Kimi-K2-Thinking VRAM & Memory Requirements

Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Unsloth): 579.24 GiB–680.01 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.

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

Publisher and provenance

PublisherMoonshot AI
Hugging Facemoonshotai/Kimi-K2-Thinking
Revisiona51ccc050d73dab088bf7b0e2dd9b30ae85a4e55
Total parameters1,026,408,240,256
Active parametersUnknown / not independently sourced
Layers61
Query heads64
KV heads64 (mla)
Head dimension112
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: 578.58 GiB (621,245,226,208 bytes)
Raw KV cache: 0.54 GiB (575,668,224 bytes)
Engine extras / workspace: 0.13 GiB–100.89 GiB
Total estimated demand: 579.24 GiB–680.01 GiB

Published artifacts

Quantization	Selected files	Bytes	Engine	Evidence
BF16 safetensors bf16 Calculator: BF16 safetensors	62 safetensors shards Total selected bytes: 594,205,920,512 (553.40 GiB / 594.21 GB) First shard: model-00001-of-000062.safetensors Last shard: model-00062-of-000062.safetensors View all 62 shard files `model-00001-of-000062.safetensors` `model-00002-of-000062.safetensors` `model-00003-of-000062.safetensors` `model-00004-of-000062.safetensors` `model-00005-of-000062.safetensors` `model-00006-of-000062.safetensors` `model-00007-of-000062.safetensors` `model-00008-of-000062.safetensors` `model-00009-of-000062.safetensors` `model-00010-of-000062.safetensors` `model-00011-of-000062.safetensors` `model-00012-of-000062.safetensors` `model-00013-of-000062.safetensors` `model-00014-of-000062.safetensors` `model-00015-of-000062.safetensors` `model-00016-of-000062.safetensors` `model-00017-of-000062.safetensors` `model-00018-of-000062.safetensors` `model-00019-of-000062.safetensors` `model-00020-of-000062.safetensors` `model-00021-of-000062.safetensors` `model-00022-of-000062.safetensors` `model-00023-of-000062.safetensors` `model-00024-of-000062.safetensors` `model-00025-of-000062.safetensors` `model-00026-of-000062.safetensors` `model-00027-of-000062.safetensors` `model-00028-of-000062.safetensors` `model-00029-of-000062.safetensors` `model-00030-of-000062.safetensors` `model-00031-of-000062.safetensors` `model-00032-of-000062.safetensors` `model-00033-of-000062.safetensors` `model-00034-of-000062.safetensors` `model-00035-of-000062.safetensors` `model-00036-of-000062.safetensors` `model-00037-of-000062.safetensors` `model-00038-of-000062.safetensors` `model-00039-of-000062.safetensors` `model-00040-of-000062.safetensors` `model-00041-of-000062.safetensors` `model-00042-of-000062.safetensors` `model-00043-of-000062.safetensors` `model-00044-of-000062.safetensors` `model-00045-of-000062.safetensors` `model-00046-of-000062.safetensors` `model-00047-of-000062.safetensors` `model-00048-of-000062.safetensors` `model-00049-of-000062.safetensors` `model-00050-of-000062.safetensors` `model-00051-of-000062.safetensors` `model-00052-of-000062.safetensors` `model-00053-of-000062.safetensors` `model-00054-of-000062.safetensors` `model-00055-of-000062.safetensors` `model-00056-of-000062.safetensors` `model-00057-of-000062.safetensors` `model-00058-of-000062.safetensors` `model-00059-of-000062.safetensors` `model-00060-of-000062.safetensors` `model-00061-of-000062.safetensors` `model-00062-of-000062.safetensors`	594,205,920,512 (553.40 GiB / 594.21 GB)	`vllm, sglang, unspecified`	`file-metadata`
GGUF Q4_K_M (Unsloth) gguf-q4_k_m Calculator: GGUF Q4_K_M (Unsloth)	13 safetensors shards Total selected bytes: 621,245,226,208 (578.58 GiB / 621.25 GB) First shard: Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00001-of-00013.gguf Last shard: Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00013-of-00013.gguf View all 13 shard files `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00001-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00002-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00003-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00004-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00005-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00006-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00007-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00008-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00009-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00010-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00011-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00012-of-00013.gguf` `Q4_K_M/Kimi-K2-Thinking-Q4_K_M-00013-of-00013.gguf`	621,245,226,208 (578.58 GiB / 621.25 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q5_K_M (Unsloth) gguf-q5_k_m Calculator: GGUF Q5_K_M (Unsloth)	16 safetensors shards Total selected bytes: 728,735,735,328 (678.69 GiB / 728.74 GB) First shard: Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00001-of-00016.gguf Last shard: Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00016-of-00016.gguf View all 16 shard files `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00001-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00002-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00003-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00004-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00005-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00006-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00007-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00008-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00009-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00010-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00011-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00012-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00013-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00014-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00015-of-00016.gguf` `Q5_K_M/Kimi-K2-Thinking-Q5_K_M-00016-of-00016.gguf`	728,735,735,328 (678.69 GiB / 728.74 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q6_K (Unsloth) gguf-q6_k Calculator: GGUF Q6_K (Unsloth)	18 safetensors shards Total selected bytes: 842,909,134,624 (785.02 GiB / 842.91 GB) First shard: Q6_K/Kimi-K2-Thinking-Q6_K-00001-of-00018.gguf Last shard: Q6_K/Kimi-K2-Thinking-Q6_K-00018-of-00018.gguf View all 18 shard files `Q6_K/Kimi-K2-Thinking-Q6_K-00001-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00002-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00003-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00004-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00005-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00006-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00007-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00008-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00009-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00010-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00011-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00012-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00013-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00014-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00015-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00016-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00017-of-00018.gguf` `Q6_K/Kimi-K2-Thinking-Q6_K-00018-of-00018.gguf`	842,909,134,624 (785.02 GiB / 842.91 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q8_0 (Unsloth) gguf-q8_0 Calculator: GGUF Q8_0 (Unsloth)	23 safetensors shards Total selected bytes: 1,091,053,887,872 (1,016.12 GiB / 1,091.05 GB) First shard: Q8_0/Kimi-K2-Thinking-Q8_0-00001-of-00023.gguf Last shard: Q8_0/Kimi-K2-Thinking-Q8_0-00023-of-00023.gguf View all 23 shard files `Q8_0/Kimi-K2-Thinking-Q8_0-00001-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00002-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00003-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00004-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00005-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00006-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00007-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00008-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00009-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00010-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00011-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00012-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00013-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00014-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00015-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00016-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00017-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00018-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00019-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00020-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00021-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00022-of-00023.gguf` `Q8_0/Kimi-K2-Thinking-Q8_0-00023-of-00023.gguf`	1,091,053,887,872 (1,016.12 GiB / 1,091.05 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
INT4 theoretical	No published file	513,204,120,128 (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	554.16 GiB–639.46 GiB	554.56 GiB–640.01 GiB	556.17 GiB–642.20 GiB
GGUF Q4_K_M (Unsloth)	578.84 GiB–679.43 GiB	579.24 GiB–680.01 GiB	580.85 GiB–682.28 GiB
GGUF Q5_K_M (Unsloth)	678.95 GiB–796.86 GiB	679.35 GiB–797.43 GiB	680.96 GiB–799.71 GiB
GGUF Q6_K (Unsloth)	785.28 GiB–921.58 GiB	785.68 GiB–922.16 GiB	787.29 GiB–924.44 GiB
GGUF Q8_0 (Unsloth)	1,016.38 GiB–1,192.67 GiB	1,016.78 GiB–1,193.25 GiB	1,018.39 GiB–1,195.52 GiB
INT4 theoretical Theoretical rows are not downloadable artifacts.	478.22 GiB–616.37 GiB	478.62 GiB–616.94 GiB	480.23 GiB–619.22 GiB

Raw KV contribution

Open the calculator with this model

Sources

Kimi-K2-Thinking model card
config.json — revision a51ccc050d73dab088bf7b0e2dd9b30ae85a4e55
safetensors index — metadata.total_size=594205920512
Artifact repo unsloth/Kimi-K2-Thinking-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.
Weight bytes come from 13 published safetensors shards totaling 621245226208 bytes.
Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
Concurrent sequences share one model replica; weights are not multiplied by sequence count.

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.

Kimi-K2-Thinking VRAM & Memory Requirements

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