Guides / 8 GB GPU / laptop

Best LLMs that run on an 8 GB GPU or laptop

Updated November 2026

As of November 2026, the best option is Nanbeige4.2 3B by Nanbeige (Kanzhun), with a Small LM Index of 70.4, followed by Spark-X2.6 4B (67.5) and Spark-X2.5 4B (65.5).

Models that fit in an 8 GB graphics card at 4-bit quantization with an 8K-token context (weights, context cache and runtime buffers), ranked by the Small LM Index. Try other hardware and context lengths in the Can I run it? tool.

#	Model	Lab	Small LM Index	Params	Memory (4-bit)	API / 1M	License
1	Nanbeige4.2 3B	Nanbeige (Kanzhun)	70.4	4.2B	~3.8 GB	self-host	Apache-2.0
2	Spark-X2.6 4B	SparkLLM (XHToken)	67.5	4.1B	~3.4 GB	self-host	Apache-2.0
3	Spark-X2.5 4B	SparkLLM (XHToken)	65.5	4.1B	~3.4 GB	self-host	Apache-2.0
4	Qwen3.5 9B	Alibaba (Qwen)	63.9	9B	~6.4 GB	$0.028	Apache-2.0
5	Granite 4.2 8B	IBM (Granite)	63.5	8B	~6.9 GB	$0.082	Apache-2.0
6	LFM3 1.7B	Liquid AI	60.0	1.7B	~1.7 GB	self-host	LFM Open License
7	Nemotron Labs Dory 1.0 4B	NVIDIA	60.0	4B	~3.4 GB	self-host	OpenMDW-1.1
8	Qwen3.5 4B	Alibaba (Qwen)	57.8	4B	~3.3 GB	self-host	Apache-2.0
9	MiniCPM5 2B	OpenBMB (MiniCPM)	57.5	2.5B	~2.4 GB	self-host	Apache-2.0
10	Falcon H1R 7B	TII (Falcon)	57.5	7.6B	~5.6 GB	self-host	Falcon LLM License
11	Qwen3 4B Thinking 2507	Alibaba (Qwen)	55.7	4B	~4.2 GB	self-host	Apache-2.0
12	Ministral 3 8B	Mistral AI	55.2	8B	~6.7 GB	$0.15	Apache-2.0
13	Granite 4.2 3B	IBM (Granite)	54.8	3B	~3 GB	self-host	Apache-2.0
14	DeepSeek-R1-0528-Qwen3-8B	DeepSeek	53.1	8.2B	~6.8 GB	self-host	MIT
15	MiniCPM-SALA	OpenBMB (MiniCPM)	52.6	9.5B	~6.7 GB	self-host	Apache-2.0

Memory covers the weights only; long contexts need more. Scores are mostly lab-reported, so gaps of 1–2 points are noise. See the methodology, or filter the full leaderboard yourself.

Head to head

Questions

What is the best LLM that runs on 8 GB of VRAM?

As of November 2026, the best option is Nanbeige4.2 3B by Nanbeige (Kanzhun), with a Small LM Index of 70.4, followed by Spark-X2.6 4B (67.5) and Spark-X2.5 4B (65.5).

How is this list ranked?

Models that fit in an 8 GB graphics card at 4-bit quantization with an 8K-token context (weights, context cache and runtime buffers), ranked by the Small LM Index.

Which of the top five needs the least memory?

Spark-X2.6 4B, at about 3.4 GB with 4-bit quantization.

Which is the cheapest to use through an API?

Qwen3.5 9B, at $0.028 per million tokens from the cheapest provider on OpenRouter.