Model Intelligence

Choose local AI models with real hardware context, not catalog chaos.

Name: VRAM Check Model Registry
Creator: VRAM Check

VRAM Check turns a raw model registry into a decision surface: what is practical, what is frontier-scale, and what really fits local hardware without guesswork. We do not host or distribute models. We surface trusted technical data and official sources.

Jump to catalogCurated envelopes / official sources / practical fit first

Start with the three signals that shrink the search space fastestFind the smallest practical fit, the newest credible addition, and the real frontier ceiling.

These anchors help you decide where to start before you dive into the full catalog.

Smallest practical fitall-minilm

0.3 GB minimum VRAM / Q4_K_M

Newest credible additionDeepSeek V3.2

2026-02-20 / Deepseek / General chat / Math / reasoning

Catalog coverage121 models / 26 families

Registry last verified on 2026-03-20. Official sources and hardware-fit envelopes only.

121Catalog nowCurated live entries

26Family coverageDistinct model families

6Open-license poolFastest shortlist for practical local use

Last registry update: 2026-03-20

121 models match this current pass

Refine the catalog

Start with fit, then narrow the shortlist.

Use VRAM and use case first when you are still exploring. Use family when you already know the line you want to inspect more closely.

Showing the full catalog right now

#1Lead pick

Deepseek

NEW

DeepSeek V3.2

236B - 131,072 ctx - 2026-02-20

Latest DeepSeek V3.2 profile with improved quality and efficiency.

Starts at81 GB

Feels right at81-137 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.estimated hardware fit with custom license and 1 linked source.

General ChatMath / Reasoning

Estimated1 source

Open dossier

#2Featured

GPT OSS

NEW

GPT OSS

20B - 131,072 ctx - 2026-02-18

Open GPT-style profile in Ollama catalog for modern local experimentation.

Starts at12 GB

Feels right at12-21 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.estimated hardware fit with open license and 1 linked source.

General ChatMath / Reasoning+1 more

Estimated1 source

Open dossier

#3Featured

Kimi

NEW

Kimi K2.5

120B - 131,072 ctx - 2026-02-12

Kimi K2.5 refreshed profile with improved instruction quality.

Starts at41 GB

Feels right at41-70 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.estimated hardware fit with custom license and 1 linked source.

General ChatMath / Reasoning

Estimated1 source

Open dossier

#4Featured

Gemma

NEW

Gemma 3N

5B - 131,072 ctx - 2026-02-05

Gemma 3N profile for edge-friendly, modern local assistant workflows.

Starts at3 GB

Feels right at3-6 GB

Default quantQ4_K_M

Best starting point for general chat.estimated hardware fit with restricted and 1 linked source.

General Chat

Estimated1 source

Open dossier

#5Featured

GLM

NEW

GLM-5

32B - 131,072 ctx - 2026-01-28

GLM-5 profile for modern multilingual and reasoning workloads.

Starts at20 GB

Feels right at20-34 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.estimated hardware fit with custom license and 1 linked source.

General ChatMath / Reasoning

Estimated1 source

Open dossier

#6Tracked

Phi

Phi-4 Reasoning

14B - 32,768 ctx - 2026-01-15

Phi-4 reasoning profile tuned for analytical and math-heavy prompts.

Starts at9 GB

Feels right at9-15 GB

Default quantQ4_K_M

Best starting point for math / reasoning / general chat.community hardware fit with custom license and 1 linked source.

Math / ReasoningGeneral Chat

Community1 source

Open dossier

#7Tracked

Mistral

Devstral 2

24B - 65,536 ctx - 2025-12-01

Coding-oriented Devstral profile for software engineering workflows.

Starts at15 GB

Feels right at15-26 GB

Default quantQ4_K_M

Best starting point for coding / general chat.community hardware fit with custom license and 1 linked source.

CodingGeneral Chat

Community1 source

Open dossier

#8Tracked

Granite

Granite 4

34B - 131,072 ctx - 2025-11-30

IBM Granite 4 profile for enterprise assistant and coding tasks.

Starts at21 GB

Feels right at21-36 GB

Default quantQ4_K_M

Best starting point for general chat / coding.estimated hardware fit with custom license and 1 linked source.

General ChatCoding

Estimated1 source

Open dossier

#9Tracked

Llama

Llama 4 Maverick

400B - 131,072 ctx - 2025-11-15

Llama 4 Maverick profile for frontier-scale deployments.

Starts at136 GB

Feels right at136-232 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.estimated hardware fit with custom license and 1 linked source.

General ChatMath / Reasoning

Estimated1 source

Open dossier

#10Tracked

Llama

Llama 4 Scout

109B - 131,072 ctx - 2025-11-15

Llama 4 Scout profile focused on balanced frontier quality.

Starts at66 GB

Feels right at66-115 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.estimated hardware fit with custom license and 1 linked source.

General ChatMath / Reasoning

Estimated1 source

Open dossier

#11Tracked

Kimi

Kimi K2

120B - 131,072 ctx - 2025-11-10

Kimi K2 profile for high-end local reasoning and assistant tasks.

Starts at41 GB

Feels right at41-70 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.estimated hardware fit with custom license and 1 linked source.

General ChatMath / Reasoning

Estimated1 source

Open dossier

#12Tracked

Qwen

Qwen3.5 32B Instruct

32B - 131,072 ctx - 2025-11-02

Refined Qwen3.5 generation with stronger instruction reliability.

Starts at20 GB

Feels right at20-34 GB

Default quantQ4_K_M

Best starting point for general chat / math / reasoning.community hardware fit with custom license and 1 linked source.

General ChatMath / Reasoning

Community1 source

Open dossier