Dataset·Catalogue

Local Models by Memory Tier

Name: Local Models by Memory Tier
Creator: Hermosa AI
Published: 2026-08-22
License: https://creativecommons.org/licenses/by/4.0/

The best open-weight model to run at each shared-memory spec, from phone-class SLMs to frontier-scale MoE quants. Every pick carries a recommended quant, a verified footprint, and why it beat the field. Browse it below, or take the raw files.

Snapshot: 2026-08-22
Coverage: 20models·18current·7of 7 tiers·14MoE
Download

About this dataset

[show]

Unit: One row = one model release with its recommended quant: the developer, architecture, parameter count, context length, the runtimes it loads on, and a why-best case for why it is the strongest pick at that memory tier and focus. Superseded picks are kept for history rather than removed.
Grouping: Rows are grouped by memory tier, the smallest shared-memory spec the recommended quant fits comfortably in, from 8GB phones and base laptops up to 512GB workstations and multi-GPU rigs.
Fit rule: A model is assigned to a tier when its quant weights plus a usable KV cache sit within roughly 75 percent of that tier’s RAM, leaving headroom for the OS and other apps.
Focus & architecture: Focus tracks what a pick is best at (general, coding, or reasoning); architecture separates dense checkpoints from mixture-of-experts models, where only a fraction of parameters activate per token.
Sources: Each row is verified against the model’s own card, its GGUF (or native-format) file listing for the quant size, and the vendor’s own benchmark tables. URLs travel with the dataset.
License: Free to use, distribute, and reproduce with attribution to Hermosa Labs LLC (CC BY 4.0).
Cite: Hermosa Labs LLC, “Local Models by Memory Tier,” 2026-08-22.

The catalogue

View

Focus

Status

20 of 20 models

#
1	DeepSeek-V4-FlashDeepSeek AI [General][Current][MoE]	2026-07-31	284B	13B	155.1GB
2	Gemma 3 12BGoogle DeepMind [General][Current][Dense]	2025-03-12	12.2B	?	7.3GB
3	GLM-4.5-AirZhipu AI (Z.ai) [General][Current][MoE]	2025-07-28	106B	12B	73GB
4	GLM-5.2Z.ai (Zhipu AI) [General][Superseded][MoE]	2026-06-17	753B	40B	253.9GB
5	GLM-5.3Z.ai (Zhipu AI) [Coding][Current][MoE]	2026-08-28	753B	40B	343GB
6	gpt-oss-120bOpenAI [Reasoning][Current][MoE]	2025-08-05	117B	5.1B	60.6GB
7	K2-Horizon-MoVA-36B-A4BInstitute of Foundation Models (MBZUAI) [General][Current][MoE]	2026-09-03	36B	4B	22.4GB
8	MiniMax-M2.7MiniMax [Coding][Current][MoE]	2026-03-18	230B	10B	140.8GB
9	Qwen3-14BAlibaba Qwen [Reasoning][Current][Dense]	2025-04-29	14.8B	?	9GB
10	Qwen3-235B-A22B-Thinking-2507Alibaba Qwen [Reasoning][Current][MoE]	2025-07-25	235B	22B	142.2GB
11	Qwen3-30B-A3BAlibaba Qwen [General][Superseded][MoE]	2025-04-29	30.5B	3.3B	18.6GB
12	Qwen3-30B-A3B-Instruct-2507Alibaba Qwen [General][Current][MoE]	2025-07-29	30.5B	3.3B	18.6GB
13	Qwen3-30B-A3B-Thinking-2507Alibaba Qwen [Reasoning][Current][MoE]	2025-07-31	30.5B	3.3B	18.6GB
14	Qwen3-4B-Instruct-2507Alibaba Qwen [General][Current][Dense]	2025-08-06	4B	?	2.5GB
15	Qwen3-4B-Thinking-2507Alibaba Qwen [Reasoning][Current][Dense]	2025-08-06	4B	?	2.5GB
16	Qwen3-Coder-30B-A3B-InstructAlibaba Qwen [Coding][Current][MoE]	2025-07-31	30.5B	3.3B	18.6GB
17	Qwen3-Coder-480B-A35B-InstructAlibaba Qwen [Coding][Current][MoE]	2025-07-22	480B	35B	290.1GB
18	Qwen3-Coder-NextAlibaba Qwen [Coding][Current][MoE]	2026-02-04	80B	3B	36.7GB
19	Seed-Coder-8B-InstructByteDance Seed [Coding][Current][Dense]	2025-06-04	8B	?	5.07GB
20	Seed-OSS-36B-InstructByteDance Seed [Reasoning][Current][Dense]	2025-08-20	36B	?	21.8GB
		median	36(20/20)	7.55(14/20)	22.1(20/20)

Local Models by Memory Tier

Snapshot

2026-08-22

Coverage

20models·18current·7of 7 tiers·14MoE

Download

DeepSeek-V4-FlashDeepSeek AI

[General][Current][MoE]

2026-07-31

284B

13B

155.1GB

Gemma 3 12BGoogle DeepMind

[General][Current][Dense]

2025-03-12

12.2B

7.3GB

GLM-4.5-AirZhipu AI (Z.ai)

[General][Current][MoE]

2025-07-28

106B

12B

73GB

GLM-5.2Z.ai (Zhipu AI)

[General][Superseded][MoE]

2026-06-17

753B

40B

253.9GB

GLM-5.3Z.ai (Zhipu AI)

[Coding][Current][MoE]

2026-08-28

753B

40B

343GB

gpt-oss-120bOpenAI

[Reasoning][Current][MoE]

2025-08-05

117B

5.1B

60.6GB

K2-Horizon-MoVA-36B-A4BInstitute of Foundation Models (MBZUAI)

[General][Current][MoE]

2026-09-03

36B

22.4GB

MiniMax-M2.7MiniMax

[Coding][Current][MoE]

2026-03-18

230B

10B

140.8GB

Qwen3-14BAlibaba Qwen

[Reasoning][Current][Dense]

2025-04-29

14.8B

9GB

Qwen3-235B-A22B-Thinking-2507Alibaba Qwen

[Reasoning][Current][MoE]

2025-07-25

235B

22B

142.2GB

Qwen3-30B-A3BAlibaba Qwen

[General][Superseded][MoE]

2025-04-29

30.5B

3.3B

18.6GB

Qwen3-30B-A3B-Instruct-2507Alibaba Qwen

[General][Current][MoE]

2025-07-29

30.5B

3.3B

18.6GB

Qwen3-30B-A3B-Thinking-2507Alibaba Qwen

[Reasoning][Current][MoE]

2025-07-31

30.5B

3.3B

18.6GB

Qwen3-4B-Instruct-2507Alibaba Qwen

[General][Current][Dense]

2025-08-06

2.5GB

Qwen3-4B-Thinking-2507Alibaba Qwen

[Reasoning][Current][Dense]

2025-08-06

2.5GB

Qwen3-Coder-30B-A3B-InstructAlibaba Qwen

[Coding][Current][MoE]

2025-07-31

30.5B

3.3B

18.6GB

Qwen3-Coder-480B-A35B-InstructAlibaba Qwen

[Coding][Current][MoE]

2025-07-22

480B

35B

290.1GB

Qwen3-Coder-NextAlibaba Qwen

[Coding][Current][MoE]

2026-02-04

80B

36.7GB

Seed-Coder-8B-InstructByteDance Seed

[Coding][Current][Dense]

2025-06-04

5.07GB

Seed-OSS-36B-InstructByteDance Seed

[Reasoning][Current][Dense]

2025-08-20

36B

21.8GB

median

36(20/20)

7.55(14/20)

22.1(20/20)