Open-weight models that run on GeForce RTX 4090

Memory 24 GB
Bandwidth 1008 GB/s
FP16 dense 165.2 TFLOPS
Usable 21.6 GB

Specs: Memory · Compare Current and Previous GeForce Series of Graphics Cards | NVIDIA · Bandwidth · NVIDIA Ada GPU Architecture (whitepaper V2.02) · FP16 dense · NVIDIA Ada GPU Architecture (whitepaper V2.02)

Search Precision Context Concurrent requests Minimum speed Architecture Type Maker Sort

Fit	Model	Parameters / active	Precision	Memory	Est. speed	Longest context	Runs as
Runs well	Qwen3.5-35B-A3B Alibaba Qwen	36B / 3B	IQ4_XS	18.8 / 21.6 GB	286–429 tok/s	149K	One GPU
Runs well	Qwen3.6-35B-A3B Alibaba Qwen	36B / 3B	IQ4_XS	18.8 / 21.6 GB	286–429 tok/s	149K	One GPU
Runs well	Qwen2.5-VL-32B-Instruct Alibaba Qwen	33.5B	IQ4_XS	19.4 / 21.6 GB	25–38 tok/s	17K	One GPU
Runs well	Qwen3-VL-32B-Instruct Alibaba Qwen	33.4B	IQ4_XS	19.3 / 21.6 GB	25–38 tok/s	17K	One GPU
Runs well	Qwen3-VL-32B-Thinking Alibaba Qwen	33.4B	IQ4_XS	19.3 / 21.6 GB	25–38 tok/s	17K	One GPU
Runs well	Qwen2.5-32B-Instruct Alibaba Qwen	32.8B	IQ4_XS	19.0 / 21.6 GB	26–39 tok/s	18K	One GPU
Runs well	Qwen2.5-Coder-32B-Instruct Alibaba Qwen	32.8B	IQ4_XS	19.0 / 21.6 GB	26–39 tok/s	18K	One GPU
Runs well	QwQ-32B Alibaba Qwen	32.8B	IQ4_XS	19.0 / 21.6 GB	26–39 tok/s	18K	One GPU
Runs well	DeepSeek-R1-Distill-Qwen-32B DeepSeek	32.8B	IQ4_XS	19.0 / 21.6 GB	26–39 tok/s	18K	One GPU
Runs well	Qwen3-32B Alibaba Qwen	32.8B	IQ4_XS	19.0 / 21.6 GB	26–39 tok/s	18K	One GPU
Runs well	Qwen3-Omni-30B-A3B-Captioner Alibaba Qwen	31.7B	Q4_0	18.2 / 21.6 GB	150–225 tok/s	44K	One GPU
Runs well	Gemma 4 31B IT Google	31.3B	Q4_0	19.2 / 21.6 GB	25–38 tok/s	23K	One GPU
Runs well	GLM-4.7-Flash Zhipu AI	31.2B	Q4_K_M	19.0 / 21.6 GB	140–211 tok/s	59K	One GPU
Runs well	Qwen3-VL-30B-A3B-Instruct Alibaba Qwen	31.1B	Q4_K_Mmaker’s file	18.9 / 21.6 GB	161–242 tok/s	37K	One GPU
Runs well	Qwen3-VL-30B-A3B-Thinking Alibaba Qwen	31.1B	Q4_K_Mmaker’s file	18.9 / 21.6 GB	161–242 tok/s	37K	One GPU
Runs well	Qwen3-30B-A3B-Instruct-2507 Alibaba Qwen	30.5B / 3.3B	Q4_K_M	18.9 / 21.6 GB	179–269 tok/s	37K	One GPU
Runs well	Qwen3-30B-A3B-Thinking-2507 Alibaba Qwen	30.5B / 3.3B	Q4_K_M	18.9 / 21.6 GB	179–269 tok/s	37K	One GPU
Runs well	Qwen3-30B-A3B Alibaba Qwen	30.5B / 3.3B	Q4_K_Mmaker’s file	18.9 / 21.6 GB	179–269 tok/s	37K	One GPU
Runs well	Qwen3-Coder-30B-A3B-Instruct Alibaba Qwen	30.5B / 3.3B	Q4_K_M	18.9 / 21.6 GB	179–269 tok/s	37K	One GPU
Runs well	Hy-Memory-30B-A3B Tencent	30.1B / 3B	Q4_K_Mmaker’s file	18.6 / 21.6 GB	192–288 tok/s	40K	One GPU
Runs well	Granite Switch 4.1 30B IBM	30B	IQ4_XS	17.6 / 21.6 GB	28–42 tok/s	23K	One GPU
Runs well	Qwen3.5-27B Alibaba Qwen	27.8B	Q5_0	19.2 / 21.6 GB	26–38 tok/s	46K	One GPU
Runs well	Qwen3.6-27B Alibaba Qwen	27.8B	Q5_0	19.2 / 21.6 GB	26–38 tok/s	46K	One GPU
Runs well	Qwen3.8-27B Alibaba Qwen	27.8B	Q5_0	19.2 / 21.6 GB	26–38 tok/s	46K	One GPU
Runs well	Gemma 4 26B A4B IT Google	25.8B / 3.8B	Q5_K_M	18.5 / 21.6 GB	155–233 tok/s	88K	One GPU
Runs well	Magistral Small 1.2 Mistral AI	24B	Q5_K_Mmaker’s file	17.6 / 21.6 GB	28–42 tok/s	33K	One GPU
Runs well	Devstral Small 1.0 Mistral AI	23.6B	Q5_K_M	17.6 / 21.6 GB	28–42 tok/s	33K	One GPU
Runs well	Devstral Small 1.1 Mistral AI	23.6B	Q5_K_M	17.6 / 21.6 GB	28–42 tok/s	33K	One GPU
Runs well	Magistral Small 1.0 Mistral AI	23.6B	Q5_K_M	17.6 / 21.6 GB	28–42 tok/s	33K	One GPU
Runs well	Magistral Small 1.1 Mistral AI	23.6B	Q5_K_Mmaker’s file	17.6 / 21.6 GB	28–42 tok/s	33K	One GPU
Runs well	ERNIE-4.5-21B-A3B-Thinking Baidu	21.8B / 3B	Q6_K	18.0 / 21.6 GB	172–258 tok/s	74K	One GPU
Runs well	GPT OSS 20B OpenAI	20.9B / 3.6B	MXFP4 (native)maker’s file	13.6 / 21.6 GB	196–294 tok/s	128K	One GPU
Runs well	Qwen2.5-14B-Instruct-1M Alibaba Qwen	14.8B	Q8_0	16.8 / 21.6 GB	29–44 tok/s	33K	One GPU
Runs well	Qwen2.5-14B-Instruct Alibaba Qwen	14.8B	Q8_0maker’s file	16.9 / 21.6 GB	29–44 tok/s	32K	One GPU
Runs well	DeepSeek-R1-Distill-Qwen-14B DeepSeek	14.8B	Q8_0	16.8 / 21.6 GB	29–44 tok/s	32K	One GPU
Runs well	Qwen3-14B Alibaba Qwen	14.8B	Q8_0maker’s file	16.6 / 21.6 GB	30–44 tok/s	40K	One GPU
Runs well	Gemma 3 12B IT Google	12.2B	Q8_0	15.5 / 21.6 GB	32–47 tok/s	25K	One GPU
Runs well	Qwen2.5-Omni-7B Alibaba Qwen	10.7B	Q8_0	11.6 / 21.6 GB	42–64 tok/s	32K	One GPU
Runs well	GLM-4.6V-Flash Zhipu AI	10.3B	Q8_0	11.0 / 21.6 GB	45–67 tok/s	125K	One GPU
Runs well	Qwen3.5-9B Alibaba Qwen	9.7B	BF16 (native)maker’s file	19.1 / 21.6 GB	26–39 tok/s	87K	One GPU
Runs well	Qianfan-VL-8B Baidu	8.8B	BF16 (native)maker’s file	18.2 / 21.6 GB	27–40 tok/s	32K	One GPU
Runs well	Granite Switch 4.1 8B IBM	8.8B	BF16 (native)maker’s file	18.5 / 21.6 GB	27–40 tok/s	27K	One GPU
Runs well	Qwen3-VL-8B-Instruct Alibaba Qwen	8.8B	BF16 (native)maker’s file	18.3 / 21.6 GB	27–40 tok/s	31K	One GPU
Runs well	Qwen3-VL-8B-Thinking Alibaba Qwen	8.8B	BF16 (native)maker’s file	18.3 / 21.6 GB	27–40 tok/s	31K	One GPU
Runs well	pplx-embed-v2-context-9b Perplexity	8.4B	Q8_0	9.1 / 21.6 GB	55–82 tok/s	256K	One GPU
Runs well	Qwen2.5-VL-7B-Instruct Alibaba Qwen	8.3B	BF16 (native)maker’s file	16.7 / 21.6 GB	30–44 tok/s	98K	One GPU
Runs well	Qwen3-8B Alibaba Qwen	8.2B	BF16 (native)maker’s file	17.1 / 21.6 GB	29–43 tok/s	39K	One GPU
Runs well	DeepSeek-R1-Distill-Llama-8B DeepSeek	8B	BF16 (native)maker’s file	16.7 / 21.6 GB	29–44 tok/s	47K	One GPU
Runs well	Gemma 4 E4B Google	8B	BF16 (native)maker’s file	15.8 / 21.6 GB	31–47 tok/s	125K	One GPU
Runs well	Ling-3.0-tiny Ant Group	7.9B / 1.3B	BF16 (native)maker’s file	15.6 / 21.6 GB	178–267 tok/s	128K	One GPU
Runs well	Qwen2.5-7B-Instruct-1M Alibaba Qwen	7.6B	BF16 (native)maker’s file	15.3 / 21.6 GB	32–48 tok/s	122K	One GPU
Runs well	Qwen2.5-7B-Instruct Alibaba Qwen	7.6B	BF16 (native)maker’s file	15.3 / 21.6 GB	32–48 tok/s	32K	One GPU
Runs well	DeepSeek-R1-Distill-Qwen-7B DeepSeek	7.6B	BF16 (native)maker’s file	15.3 / 21.6 GB	32–48 tok/s	32K	One GPU
Runs well	Mistral-7B-Instruct-v0.1 Mistral AI	7.2B	BF16 (native)maker’s file	15.2 / 21.6 GB	32–49 tok/s	32K	One GPU
Runs well	Gemma 4 E2B Google	5.1B	BF16 (native)maker’s file	10.1 / 21.6 GB	49–73 tok/s	125K	One GPU
Runs well	WeVisOne-VL-4B Tencent	4.8B	BF16 (native)maker’s file	10.6 / 21.6 GB	46–70 tok/s	86K	One GPU
Runs well	Qianfan-OCR Baidu	4.7B	BF16 (native)maker’s file	10.5 / 21.6 GB	47–71 tok/s	31K	One GPU
Runs well	Qwen3.5-4B Alibaba Qwen	4.7B	BF16 (native)maker’s file	9.4 / 21.6 GB	53–79 tok/s	256K	One GPU
Runs well	Qwen3-VL-4B-Instruct Alibaba Qwen	4.4B	BF16 (native)maker’s file	9.9 / 21.6 GB	50–75 tok/s	91K	One GPU
Runs well	Qwen3-4B-Instruct-2507 Alibaba Qwen	4B	BF16 (native)maker’s file	9.1 / 21.6 GB	54–82 tok/s	97K	One GPU
Runs well	Qwen3-4B Alibaba Qwen	4B	BF16 (native)maker’s file	9.1 / 21.6 GB	54–82 tok/s	128K	One GPU
Runs well	DeepSeek-OCR DeepSeek	3.3B	BF16 (native)maker’s file	7.2 / 21.6 GB	181–271 tok/s	8K	One GPU
Runs well	Granite-4.0-H-Micro IBM	3.2B	BF16 (native)maker’s file	6.5 / 21.6 GB	78–117 tok/s	128K	One GPU
Runs well	Qwen3-1.7B Alibaba Qwen	2B	BF16 (native)maker’s file	5.2 / 21.6 GB	101–151 tok/s	40K	One GPU
Runs well	DeepSeek-R1-Distill-Qwen-1.5B DeepSeek	1.8B	BF16 (native)maker’s file	4.0 / 21.6 GB	133–200 tok/s	125K	One GPU
Runs well	GLM-OCR Zhipu AI	1.3B	BF16 (native)maker’s file	3.5 / 21.6 GB	158–237 tok/s	128K	One GPU
Runs well	Qwen3-0.6B Alibaba Qwen	0.8B	BF16 (native)maker’s file	2.8 / 21.6 GB	206–309 tok/s	40K	One GPU
Runs well	EmbeddingGemma 2 Google	0.7B	BF16 (native)maker’s file	2.0 / 21.6 GB	320–479 tok/s	256K	One GPU
Runs well	Qwen3-Embedding-0.6B Alibaba Qwen	0.6B	BF16 (native)maker’s file	2.5 / 21.6 GB	236–355 tok/s	32K	One GPU
Runs well	bge-reranker-v2-m3 BAAI	0.6B	FP32 (native)maker’s file	3.4 / 21.6 GB	164–246 tok/s	8K	One GPU
Runs well	bge-m3-unsupervised BAAI	0.6B	FP32 (native)maker’s file	3.4 / 21.6 GB	164–246 tok/s	8K	One GPU
Runs well	bge-large-en-v1.5 BAAI	0.3B	FP32 (native)maker’s file	1.8 / 21.6 GB	362–544 tok/s	512	One GPU
Runs well	bge-reranker-base BAAI	0.3B	FP32 (native)maker’s file	1.6 / 21.6 GB	446–668 tok/s	514	One GPU
Runs well	bge-base-en-v1.5 BAAI	0.1B	FP32 (native)maker’s file	0.9 / 21.6 GB	1103–1655 tok/s	512	One GPU
Runs well	bge-small-en-v1.5 BAAI	0B	FP32 (native)maker’s file	0.6 / 21.6 GB	3527–5290 tok/s	512	One GPU
Needs system memory	Ming-3.0-audio-flash Ant Group	128.2B / 6.5B	Q4_K_M	76.5 GB, 55.3 in system memory	14–22 tok/s	—	32/42 layers in system memory
Needs system memory	Ling-3.0-flash Ant Group	127.5B / 5.1B	Q4_K_Mmaker’s file	75.7 GB, 54.6 in system memory	19–28 tok/s	—	32/42 layers in system memory
Needs system memory	Qwen3.5-122B-A10B Alibaba Qwen	125.1B / 10B	IQ4_XS	65.2 GB, 43.8 in system memory	11–17 tok/s	—	34/48 layers in system memory
Needs system memory	Ling-3.0-flash-VL Ant Group	124.8B / 5.5B	Q4_K_M	74.5 GB, 53.8 in system memory	17–26 tok/s	—	32/42 layers in system memory
Needs system memory	gpt-oss-120b OpenAI	116.8B / 5.1B	MXFP4 (native)maker’s file	64.1 GB, 43.9 in system memory	21–32 tok/s	—	26/36 layers in system memory
Needs system memory	GLM-4.5-Air Zhipu AI	110.5B	IQ4_XS	58.8 GB, 49.3 in system memory	12–18 tok/s	—	Experts in system memory
Needs system memory	GLM-4.5V Zhipu AI	107.7B / 12B	IQ4_XS	57.4 GB, 50.5 in system memory	12–18 tok/s	—	Experts in system memory
Needs system memory	GLM-4.6V Zhipu AI	107.7B	IQ4_XS	57.4 GB, 49.3 in system memory	12–18 tok/s	—	Experts in system memory
Needs system memory	Ling-2.6-flash Ant Group	107.5B / 7.4B	Q4_K_M	64.1 GB, 43.7 in system memory	13–20 tok/s	—	23/32 layers in system memory
Needs system memory	Qwen3-Next-80B-A3B-Instruct Alibaba Qwen	81.3B / 3B	Q6_Kmaker’s file	64.3 GB, 43.2 in system memory	25–38 tok/s	—	34/48 layers in system memory
Needs system memory	Qwen3-Next-80B-A3B-Thinking Alibaba Qwen	81.3B / 3B	Q6_Kmaker’s file	64.3 GB, 43.2 in system memory	25–38 tok/s	—	34/48 layers in system memory
Needs system memory	Qwen3-Coder-Next Alibaba Qwen	79.7B / 3B	FP8maker’s file	78.8 GB, 57.7 in system memory	19–28 tok/s	—	37/48 layers in system memory
Needs system memory	Qwen2.5-VL-72B-Instruct Alibaba Qwen	73.4B	IQ4_XS	40.6 GB, 19.1 in system memory	2–3 tok/s	—	42/80 layers in system memory
Needs system memory	Qwen2-VL-72B-Instruct Alibaba Qwen	73.4B	IQ4_XS	40.6 GB, 19.1 in system memory	2–3 tok/s	—	42/80 layers in system memory
Needs system memory	Qwen2.5-72B-Instruct Alibaba Qwen	72.7B	IQ4_XS	40.3 GB, 18.9 in system memory	2–3 tok/s	—	42/80 layers in system memory
Needs system memory	Qianfan-VL-70B Baidu	71.8B	IQ4_XS	39.8 GB, 18.7 in system memory	2–3 tok/s	—	42/80 layers in system memory
Needs system memory	DeepSeek-R1-Distill-Llama-70B DeepSeek	70.6B	IQ4_XS	39.2 GB, 17.9 in system memory	2–3 tok/s	—	41/80 layers in system memory

Estimates from the makers’ Hugging Face repositories (as of 2026-11-06); filtered views aren’t indexed. How it’s estimated