Self-hosting AI models

Models with open weights, sorted by the graphics card they need at least at 4-bit – with licence, model files, best benchmark and the API price for comparison.

301models with open weights

278with a GPU memory figure

284with model files on Hugging Face

Open weights means: the model files are public, you can run the model on your own hardware. What you may do with it is governed by the licence – not every licence allows every use.

GPU memory is derived, not measured: parameters (bn) × 0.55 GB + 4 GB at 4-bit quantisation. Example: 32.8 bn × 0.55 GB + 4 GB = 22.0 GB – fits a 24 GB card. This is a lower bound: common 4-bit files of large models need about 0.6 GB per billion parameters, long contexts and many concurrent requests considerably more.

Which hardware for which purpose: Hardware for local LLMs (tech blog)

24 GB – one RTX PRO 4000

GPU memory (4-bit, derived) up to 24 GB · 189 models

Model	Type	Parameters (bn, total / active)	GPU memory (4-bit, derived)GPU memory	Licence	Model files	Best benchmark	API price: InputAPI price in	API price: OutputAPI price out
@cf/ai4bharat/indictrans2-en-indic-1B	Language models	1.12	from 5 GB	MIT	huggingface.co/ai4bharat/indictrans2-en-indic-1B	–	€0.31	€0.31
@cf/aisingapore/gemma-sea-lion-v4-27b-it	Language models	27.43	from 20 GB	Gemma Terms of Use	huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT	–	€0.31	€0.50
@cf/baai/bge-reranker-base	Language models	0.28	from 5 GB	MIT	huggingface.co/BAAI/bge-reranker-base	–	€0.0027	–
@cf/baai/bge-small-en-v1.5	Embeddings	0.033	from 5 GB	MIT	huggingface.co/BAAI/bge-small-en-v1.5	–	€0.018	–
@cf/cloudflare/clef	Language models	27.36	from 20 GB	Apache-2.0	huggingface.co/Cloudflare/clef	–	–	–
@cf/cloudflare/clef-omni	Language models	35.26 / 3	from 24 GB	Apache-2.0	huggingface.co/Cloudflare/clef-omni	–	–	–
@cf/deepseek-ai/deepseek-r1-distill-qwen-32b	Language models	32.76	from 23 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B	–	€0.45	€4.37
@cf/huggingface/distilbert-sst-2-int8	Language models	0.067	from 5 GB	Apache-2.0	huggingface.co/distilbert/distilbert-base-uncased-finetuned-sst-2-english	–	–	–
@cf/ibm-granite/granite-4.0-h-micro	Language models	3.19	from 6 GB	Apache-2.0	huggingface.co/ibm-granite/granite-4.0-h-micro	–	€0.015	€0.10
@cf/meta/llama-2-7b-chat-fp16	Language models	6.74	from 8 GB	Llama 2 Community License	huggingface.co/meta-llama/Llama-2-7b-chat-hf	–	€0.50	€5.97
@cf/meta/llama-guard-3-8b	Language models	8.03	from 9 GB	Llama 3.1 Community License	huggingface.co/meta-llama/Llama-Guard-3-8B	–	€0.43	€0.027
@cf/meta/m2m100-1.2b	Language models	1.2	from 5 GB	MIT	huggingface.co/facebook/m2m100_1.2B	–	€0.31	€0.31
@cf/microsoft/resnet-50	Image generation	0.026	from 5 GB	Apache-2.0	huggingface.co/microsoft/resnet-50	–	–	–
@cf/mistral/mistral-7b-instruct-v0.1	Language models	7.24	from 8 GB	Apache-2.0	huggingface.co/mistralai/Mistral-7B-Instruct-v0.1	–	€0.099	€0.17
@cf/moondream/moondream3.1-9B-A2B	Language models	9.27 / 2	from 10 GB	Moondream Model License 1.0	huggingface.co/moondream/moondream3.1-9B-A2B	–	€0.27	€0.90
@cf/pfnet/plamo-embedding-1b	Embeddings	1.05	from 5 GB	Apache-2.0	huggingface.co/pfnet/plamo-embedding-1b	–	€0.017	–
@cf/pipecat-ai/smart-turn-v2	Language models	0.095	from 5 GB	BSD-2-Clause	huggingface.co/pipecat-ai/smart-turn-v2	–	–	–
@cf/qwen/qwen2.5-coder-32b-instruct	Language models	32.76	from 23 GB	Apache-2.0	huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct	–	€0.59	€0.90
@cf/qwen/qwq-32b	Language models	32.76	from 23 GB	Apache-2.0	huggingface.co/Qwen/QwQ-32B	–	€0.59	€0.90
All MiniLM L12 v2	Embeddings	0.033	from 5 GB	Apache-2.0	huggingface.co/sentence-transformers/all-MiniLM-L12-v2	MTEB NFCorpus · rank 23 of 29	€0.0045	–
all-MiniLM-L6-v2	Embeddings	0.022	from 5 GB	Apache-2.0	huggingface.co/sentence-transformers/all-MiniLM-L6-v2	MTEB NFCorpus · rank 24 of 29	€0.0045	–
all-mpnet-base-v2	Embeddings	0.11	from 5 GB	Apache-2.0	huggingface.co/sentence-transformers/all-mpnet-base-v2	MTEB EnglishFinance3Retrieval · rank 11 of 34	€0.0045	–
Argus-Colqwen3.5-9b-v0	Embeddings	8.82	from 9 GB	Apache-2.0	huggingface.co/DataScience-UIBK/Argus-Colqwen3.5-9b-v0	–	–	–
AutoGLM Phone 9B Multilingual	Language models	10.29	from 10 GB	MIT	huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual	–	–	–
Aya Expanse	Language models	32.3	from 22 GB	CC-BY-NC-4.0	huggingface.co/CohereLabs/aya-expanse-32b	–	–	–
bekko-embedding-v1-a25m	Embeddings	0.12	from 5 GB	MIT	huggingface.co/hotchpotch/bekko-embedding-v1-a25m	MTEB GermanQuAD-Retrieval · rank 3 of 26	–	–
bekko-embedding-v1-a8m	Embeddings	0.11	from 5 GB	MIT	huggingface.co/hotchpotch/bekko-embedding-v1-a8m	MTEB GermanQuAD-Retrieval · rank 4 of 26	–	–
BGE Multilingual Gemma2	Embeddings	9.24	from 10 GB	Gemma Terms	huggingface.co/BAAI/bge-multilingual-gemma2	MTEB ArguAna · rank 2 of 38	€0.01	–
BGE Reranker v2 M3	Language models	0.57	from 5 GB	Apache-2.0	huggingface.co/BAAI/bge-reranker-v2-m3	–	€0.011	–
bge-base-en-v1.5	Embeddings	0.11	from 5 GB	MIT	huggingface.co/BAAI/bge-base-en-v1.5	MTEB SciFact · rank 4 of 28	€0.0045	–
bge-en-icl	Embeddings	7.11	from 8 GB	Apache-2.0	huggingface.co/BAAI/bge-en-icl	–	€0.009	–
bge-large-en-v1.5	Embeddings	0.34	from 5 GB	MIT	huggingface.co/BAAI/bge-large-en-v1.5	MTEB SciFact · rank 3 of 28	€0.009	–
BGE-M3	Embeddings	0.57	from 5 GB	MIT	huggingface.co/BAAI/bge-m3	MTEB EnglishFinance1Retrieval · rank 12 of 34	€0.0082	–
Chroma	Image generation	8.9	from 9 GB	Apache-2.0	huggingface.co/lodestones/Chroma1-HD	–	–	–
Cloudflare/clef-flash	Language models	9.41	from 10 GB	Apache-2.0	huggingface.co/Cloudflare/clef-flash	–	–	–
CogView-4	Image generation	6.37	from 8 GB	Apache-2.0	huggingface.co/zai-org/CogView4-6B	–	–	–
COLMODERNVBERT	Embeddings	0.25	from 5 GB	MIT	huggingface.co/ModernVBERT/colmodernvbert	–	–	–
colnomic-embed-multimodal-7b	Embeddings	7	from 8 GB	Apache-2.0	huggingface.co/nomic-ai/colnomic-embed-multimodal-7b	–	–	–
colpali-v1.3	Embeddings	2.92	from 6 GB	Apache-2.0	huggingface.co/vidore/colpali-v1.3	–	–	–
Command R 03-2024	Language models	34.98	from 24 GB	CC-BY-NC-4.0	huggingface.co/CohereLabs/c4ai-command-r-v01	MMLU-Pro (TIGER-Lab Leaderboard) · rank 40 of 42	€0.45	€1.34
convaiinnovations/laya	Language models	0.42	from 5 GB	Apache-2.0	huggingface.co/convaiinnovations/laya	–	–	–
Cydonia 24B v4.1	Language models	23.57	from 17 GB	–	huggingface.co/TheDrummer/Cydonia-24B-v4.1	–	€0.27	€0.45
DeepSeek OCR 2	Language models	3.39	from 6 GB	Apache-2.0	huggingface.co/deepseek-ai/DeepSeek-OCR-2	–	–	–
DeepSeek-OCR	Image models	3.34	from 6 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-OCR	–	€0.27	€1.07
Devstral	Language models	23.57	from 17 GB	Apache-2.0	huggingface.co/mistralai/Devstral-Small-2507	–	–	–
e5-base-v2	Embeddings	0.11	from 5 GB	MIT	huggingface.co/intfloat/e5-base-v2	MTEB TRECCOVID · rank 24 of 37	€0.0045	–
e5-large-v2	Embeddings	0.34	from 5 GB	MIT	huggingface.co/intfloat/e5-large-v2	MTEB GermanDPR · rank 11 of 22	€0.009	–
E5-Mistral-7B	Embeddings	7.11 / 7.11	from 8 GB	MIT	huggingface.co/intfloat/e5-mistral-7b-instruct	MTEB SciFact · rank 2 of 28	€0.02	–
embeddinggemma-300m	Embeddings	0.31	from 5 GB	Gemma Terms	huggingface.co/google/embeddinggemma-300m	MTEB ArguAna · rank 4 of 38	–	–
EuroLLM 22B	Language models	22.64	from 17 GB	Apache-2.0	huggingface.co/utter-project/EuroLLM-22B-Instruct-2512	–	–	–
Flux schnell	Image generation	11.89	from 11 GB	Apache-2.0	huggingface.co/black-forest-labs/FLUX.1-schnell	–	–	–
FLUX-1-dev	Image generation	11.9	from 11 GB	FLUX.1 [dev] Non-Commercial License	huggingface.co/black-forest-labs/FLUX.1-dev	–	–	–
FLUX-2-klein-4b	Image generation	3.88	from 7 GB	Apache-2.0	huggingface.co/black-forest-labs/FLUX.2-klein-4B	–	–	–
FLUX-2-klein-9b	Image generation	9.08	from 9 GB	FLUX Non-Commercial License	huggingface.co/black-forest-labs/FLUX.2-klein-9B	–	–	–
FLUX.2 [dev]	Image generation	32.22	from 22 GB	FLUX Non-Commercial License	huggingface.co/black-forest-labs/FLUX.2-dev	–	–	–
Gemma 2 9B	Language models	9.24	from 10 GB	Gemma Terms of Use	huggingface.co/google/gemma-2-9b-it	MMLU-Pro (TIGER-Lab Leaderboard) · rank 32 of 42	–	–
Gemma 3 12B	Language models	12.19	from 11 GB	Gemma Terms of Use	huggingface.co/google/gemma-3-12b-it	Vectara HHEM-2.3 · rank 5 of 82	€0.045	€0.13
Gemma 3 27B	Language models	27.4	from 20 GB	Gemma Terms of Use	huggingface.co/google/gemma-3-27b-it	Vectara HHEM-2.3 · rank 27 of 82	€0.072	€0.14
Gemma 3 4B	Language models	4.3	from 7 GB	Gemma Terms of Use	huggingface.co/google/gemma-3-4b-it	Vectara HHEM-2.3 · rank 23 of 82	€0.018	€0.036
Gemma 4 31B	Language models	31	from 22 GB	Apache-2.0	huggingface.co/google/gemma-4-31B-it	Vectara HHEM-2.3 · rank 27 of 82	€0.09	€0.27
Gemma 4 Uncensored	Language models	25.2 / 3.8	from 18 GB	Gemma Terms of Use	huggingface.co/Jiunsong/supergemma4-26b-uncensored-gguf-v2	–	€0.15	€0.45
gemma-4-12B-it	Language models	11.95	from 11 GB	Apache-2.0	huggingface.co/google/gemma-4-12B-it	–	€0.09	€0.27
GLM 4.7 Flash	Language models	30 / 3	from 21 GB	MIT	huggingface.co/zai-org/GLM-4.7-Flash	Vectara HHEM-2.3 · rank 36 of 82	€0.031	€0.18
GLM 4.7 Flash Heretic	Language models	30 / 3	from 21 GB	MIT	huggingface.co/Olafangensan/GLM-4.7-Flash-heretic	–	€0.063	€0.36
GLM-4-32B-0414-128K	Language models	32	from 22 GB	MIT	huggingface.co/zai-org/GLM-4-32B-0414	–	€0.09	€0.09
GLM-OCR	Image models	0.9	from 5 GB	–	huggingface.co/zai-org/GLM-OCR	–	€0.027	€0.027
gpt-oss-20b	Language models	21 / 3.6	from 16 GB	Apache-2.0	huggingface.co/openai/gpt-oss-20b	MMLU-Pro (TIGER-Lab Leaderboard) · rank 16 of 42	€0.024	€0.11
Granite 4.0 H Small	Language models	32.21 / 9	from 22 GB	Apache-2.0	huggingface.co/ibm-granite/granite-4.0-h-small	Vectara HHEM-2.3 · rank 11 of 82	–	–
Granite Vision 3.2 2B	Image models	2.98	from 6 GB	Apache-2.0	huggingface.co/ibm-granite/granite-vision-3.2-2b	–	–	–
granite-embedding-278m-multilingual	Embeddings	0.28	from 5 GB	Apache-2.0	huggingface.co/ibm-granite/granite-embedding-278m-multilingual	MTEB GermanDPR · rank 10 of 22	–	–
granite-embedding-311m-multilingual-r2	Embeddings	0.31	from 5 GB	Apache-2.0	huggingface.co/ibm-granite/granite-embedding-311m-multilingual-r2	MTEB SciFact · rank 14 of 28	–	–
granite-embedding-97m-multilingual-r2	Embeddings	0.097	from 5 GB	Apache-2.0	huggingface.co/ibm-granite/granite-embedding-97m-multilingual-r2	MTEB LegalQuAD · rank 27 of 44	–	–
granite-vision-3.3-2b-embedding	Embeddings	2.98	from 6 GB	Apache-2.0	huggingface.co/ibm-granite/granite-vision-3.3-2b-embedding	–	–	–
green-embedding	Embeddings	4.02	from 7 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Embedding-4B	–	€0.20	–
green-l	Language models	24.01	from 18 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506	–	€0.25	€0.80
green-l-raw	Language models	24.01	from 18 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506	–	€0.25	€0.80
green-rerank	Language models	4.02	from 7 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Reranker-4B	–	€0.12	–
gte-base	Embeddings	0.11	from 5 GB	MIT	huggingface.co/thenlper/gte-base	MTEB GermanDPR · rank 17 of 22	€0.0045	–
gte-large	Embeddings	0.34	from 5 GB	MIT	huggingface.co/thenlper/gte-large	MTEB GermanDPR · rank 12 of 22	€0.009	–
gte-multilingual-base	Embeddings	0.3	from 5 GB	Apache-2.0	huggingface.co/Alibaba-NLP/gte-multilingual-base	MTEB SciFact · rank 7 of 28	–	–
harrier-oss-v1-270m	Embeddings	0.27 / 0.27	from 5 GB	MIT	huggingface.co/microsoft/harrier-oss-v1-270m	MTEB ArguAna · rank 9 of 38	–	–
Ideogram V4	Image generation	9.3	from 10 GB	Ideogram 4 Non-Commercial	huggingface.co/ideogram-ai/ideogram-4-fp8	–	–	–
intfloat/multilingual-e5-large	Embeddings	0.56	from 5 GB	MIT	huggingface.co/intfloat/multilingual-e5-large	MTEB GermanDPR · rank 4 of 22	€0.0082	–
jina-clip-v1	Embeddings	0.22	from 5 GB	Apache-2.0	huggingface.co/jinaai/jina-clip-v1	–	–	–
jina-clip-v2	Embeddings	0.86	from 5 GB	Apache-2.0	huggingface.co/jinaai/jina-clip-v2	–	–	–
jina-embeddings-v2-base-de	Embeddings	0.16	from 5 GB	Apache-2.0	huggingface.co/jinaai/jina-embeddings-v2-base-de	–	€0.05	–
Kokoro Text to Speech	Language models	0.082	from 5 GB	Apache-2.0	huggingface.co/hexgrad/Kokoro-82M	–	–	–
Krea 2 Turbo	Image generation	12.82	from 12 GB	Krea 2 Community License	huggingface.co/krea/Krea-2-Turbo	–	–	–
LFM2.5-VL-3B	Image models	3.12	from 6 GB	LFM Open License	huggingface.co/LiquidAI/LFM2.5-VL-3B	–	–	–
LightOnOCR 2	Image models	1.01	from 5 GB	Apache-2.0	huggingface.co/lightonai/LightOnOCR-2-1B	–	€0.15	€0.30
lightonocr-3-0.8b	Language models	0.85	from 5 GB	Apache-2.0	huggingface.co/lightonai/LightOnOCR-3-0.8B	–	–	–
Llama 3 8B	Language models	8.03	from 9 GB	Llama 3 Community License	huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 145 of 153	€0.11	€0.24
Llama 3 8B Stheno V3.2	Language models	8.03	from 9 GB	CC-BY-NC-4.0	huggingface.co/Sao10K/L3-8B-Stheno-v3.2	–	–	–
Llama 3.1 8B Instruct	Language models	8.03	from 9 GB	Llama 3.1 Community License	huggingface.co/meta-llama/Llama-3.1-8B-Instruct	MMLU-Pro (TIGER-Lab Leaderboard) · rank 36 of 42	€0.018	€0.036
Llama 3.2 1B	Language models	1.24	from 5 GB	Llama 3.2 Community License	huggingface.co/meta-llama/Llama-3.2-1B-Instruct	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 150 of 153	€0.024	€0.18
Llama 3.2 3B	Language models	3.21	from 6 GB	Llama 3.2 Community License	huggingface.co/meta-llama/Llama-3.2-3B-Instruct	Berkeley Function-Calling Leaderboard V4 · rank 40 of 42	€0.046	€0.30
Llama 3.2 Vision 11B	Image models	10.67	from 10 GB	Llama 3.2 Community License	huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct	–	€0.14	€0.14
Llama-Guard-4-12B	Language models	12	from 11 GB	Llama 4 Community License	huggingface.co/meta-llama/Llama-Guard-4-12B	–	€0.16	€0.16
LLaVA 1.5 7B	Image models	7.06	from 8 GB	Llama-2-Lizenz	huggingface.co/llava-hf/llava-1.5-7b-hf	–	–	–
M MiniCPM-V 4.5	Language models	8.7	from 9 GB	Apache-2.0	huggingface.co/openbmb/MiniCPM-V-4_5	–	€0.58	€0.98
Magistral Small 1.2	Language models	24.01	from 18 GB	Apache-2.0	huggingface.co/mistralai/Magistral-Small-2509	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 116 of 153	€0.22	€0.67
Ministral 3 14B	Image models	13.95	from 12 GB	Apache-2.0	huggingface.co/mistralai/Ministral-3-14B-Instruct-2512	Vectara HHEM-2.3 · rank 79 of 82	€0.09	€0.09
Ministral 3 3B	Language models	3.85	from 7 GB	Apache-2.0	huggingface.co/mistralai/Ministral-3-3B-Instruct-2512	Vectara HHEM-2.3 · rank 82 of 82	€0.045	€0.045
Ministral 3 8B	Language models	8.92	from 9 GB	Apache-2.0	huggingface.co/mistralai/Ministral-3-8B-Instruct-2512	searchit Entity Extraction · rank 31 of 53	€0.063	€0.063
Mistral 7B	Language models	7.25	from 8 GB	Apache-2.0	huggingface.co/mistralai/Mistral-7B-Instruct-v0.3	–	€0.13	€0.18
Mistral 7B Instruct	Language models	7.25	from 8 GB	Apache-2.0	huggingface.co/mistralai/Mistral-7B-Instruct-v0.3	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 152 of 153	€0.12	€0.12
Mistral Nemo Instruct	Language models	12.25	from 11 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Nemo-Instruct-2407	Berkeley Function-Calling Leaderboard V4 · rank 35 of 42	€0.016	€0.025
Mistral Small 24B Instruct New	Language models	24.01	from 18 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506	–	€0.099	€0.30
Mistral Small 3 24B	Language models	23.6	from 17 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501	Vectara HHEM-2.3 · rank 8 of 82	€0.045	€0.072
Mistral Small 3.1 24B	Language models	24.01	from 18 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 139 of 153	€0.09	€0.27
Mistral Small 3.2 24B	Language models	24.01	from 18 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506	Berkeley Function-Calling Leaderboard V4 · rank 26 of 42	€0.061	€0.16
mistral-7b-instruct-v0.2	Language models	7.24	from 8 GB	Apache-2.0	huggingface.co/mistralai/Mistral-7B-Instruct-v0.2	MMLU-Pro (TIGER-Lab Leaderboard) · rank 42 of 42	€0.14	€0.20
most-embed-de	Embeddings	1.14	from 5 GB	CC-BY-NC-4.0	huggingface.co/malteos/most-embed-de	MTEB LegalQuAD · rank 1 of 44	–	–
multi-qa-mpnet-base-dot-v1	Embeddings	0.11	from 5 GB	–	huggingface.co/sentence-transformers/multi-qa-mpnet-base-dot-v1	–	€0.0045	–
multilingual-e5-large-instruct	Embeddings	0.56	from 5 GB	MIT	huggingface.co/intfloat/multilingual-e5-large-instruct	MTEB GermanQuAD-Retrieval · rank 6 of 26	€0.009	–
multilingual-e5-small	Embeddings	0.12	from 5 GB	MIT	huggingface.co/intfloat/multilingual-e5-small	MTEB GermanQuAD-Retrieval · rank 10 of 26	€0.0067	–
mxbai-embed-large-v1	Embeddings	0.34	from 5 GB	Apache-2.0	huggingface.co/mixedbread-ai/mxbai-embed-large-v1	MTEB NFCorpus · rank 3 of 29	–	–
Nemotron Embed VL 1B v2	Embeddings	1.7	from 5 GB	NVIDIA Open Model License	huggingface.co/nvidia/llama-nemotron-embed-vl-1b-v2	–	€0.011	€0.011
nemotron-colembed-vl-8b-v2	Embeddings	8.77	from 9 GB	eigene Lizenz - vor Einsatz pruefen	huggingface.co/nvidia/nemotron-colembed-vl-8b-v2	–	–	–
Nemotron-Content-Safety-3.5	Language models	4	from 7 GB	OpenMDW-1.1 + Gemma Terms of Use	huggingface.co/nvidia/Nemotron-3.5-Content-Safety	–	€0.18	€0.18
nemotron-nano-v2-12b	Language models	12.31	from 11 GB	NVIDIA Open Model License	huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2	–	€0.21	€0.64
nomic-embed-text-v1.5	Embeddings	0.14	from 5 GB	Apache-2.0	huggingface.co/nomic-ai/nomic-embed-text-v1.5	MTEB SciFact · rank 16 of 28	–	–
nomic-embed-text-v2-moe	Embeddings	0.47 / 0.3	from 5 GB	Apache-2.0	huggingface.co/nomic-ai/nomic-embed-text-v2-moe	MTEB EnglishFinance1Retrieval · rank 8 of 34	–	–
nomic-embed-vision-v1.5	Embeddings	0.093	from 5 GB	Apache-2.0	huggingface.co/nomic-ai/nomic-embed-vision-v1.5	–	–	–
North Micro Vision Instruct	Image models	2.48	from 6 GB	Apache-2.0	huggingface.co/CohereLabs/north-micro-vision-instruct	–	–	–
NVIDIA Nemotron 3 Nano 30B	Language models	31.58 / 3.5	from 22 GB	NVIDIA Nemotron Open Model License	huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16	Vectara HHEM-2.3 · rank 40 of 82	€0.027	€0.11
NVIDIA Nemotron 3.5 Lightning 30B A3B	Language models	30 / 3	from 21 GB	OpenMDW-1.1	huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16	–	€0.046	€0.14
NVIDIA Nemotron Nano 2	Language models	8.89	from 9 GB	NVIDIA Open Model License	huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2	–	€0.027	€0.11
NVIDIA Nemotron Nano 2 VL	Language models	13.18	from 12 GB	NVIDIA Open Model License	huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16	–	€0.054	€0.21
Ops-Colqwen3-4B	Embeddings	4.44	from 7 GB	Apache-2.0	huggingface.co/OpenSearch-AI/Ops-Colqwen3-4B	–	–	–
Orpheus TTS	Language models	3.78	from 7 GB	Apache-2.0	huggingface.co/canopylabs/orpheus-3b-0.1-ft	–	–	–
Parakeet ASR	Language models	0.6	from 5 GB	CC-BY-4.0	huggingface.co/nvidia/parakeet-tdt-0.6b-v3	–	–	–
paraphrase-MiniLM-L6-v2	Embeddings	0.023	from 5 GB	Apache-2.0	huggingface.co/sentence-transformers/paraphrase-MiniLM-L6-v2	–	€0.0045	–
Phi-4	Language models	14.7	from 13 GB	MIT	huggingface.co/microsoft/phi-4	Vectara HHEM-2.3 · rank 3 of 82	€0.063	€0.13
Qwen 2.5 7B	Language models	7 / 7.61	from 9 GB	Apache 2.0	huggingface.co/Qwen/Qwen2.5-7B-Instruct	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 142 of 153	€0.033	€0.082
Qwen 3 TTS 0.6B	Language models	0.91	from 5 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice	–	–	–
Qwen 3 TTS 1.7B	Language models	1.92	from 6 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice	–	–	–
Qwen Image	Image generation	20	from 15 GB	Apache-2.0	huggingface.co/Qwen/Qwen-Image	–	–	–
Qwen Image 2.1 Turbo	Image generation	7.12	from 8 GB	Qwen Research License	huggingface.co/Qwen/Qwen-Image-2.1-Turbo	–	–	–
Qwen Image Edit	Image generation	20	from 15 GB	Apache-2.0	huggingface.co/Qwen/Qwen-Image-Edit	–	–	–
Qwen/Qwen3-Reranker-0.6B	Language models	0.6	from 5 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Reranker-0.6B	–	€0.0098	–
Qwen2.5-VL 32B Instruct	Image models	33.45	from 23 GB	Apache-2.0	huggingface.co/Qwen/Qwen2.5-VL-32B-Instruct	–	–	–
Qwen2.5-VL 7B Instruct	Image models	8.29	from 9 GB	Apache-2.0	huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct	–	–	–
Qwen3 30B A3B Instruct 2507	Language models	30.53 / 3.3	from 21 GB	–	huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507	Berkeley Function-Calling Leaderboard V4 · rank 22 of 42	€0.089	€0.27
Qwen3 8B	Language models	8	from 9 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-8B	Vectara HHEM-2.3 · rank 7 of 82	–	–
Qwen3 Coder 30B	Language models	30.53 / 3.3	from 21 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct	–	€0.057	€0.21
Qwen3 Embedding 8B	Embeddings	7.57 / 7.57	from 9 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Embedding-8B	MTEB TRECCOVID · rank 1 of 37	€0.009	–
Qwen3-14B	Language models	14.77	from 13 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-14B	Vectara HHEM-2.3 · rank 13 of 82	€0.11	€0.21
Qwen3-30B-A3B	Language models	30.5 / 3.3	from 21 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-30B-A3B	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 96 of 153	€0.046	€0.30
Qwen3-32B	Language models	32.8	from 23 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-32B	Vectara HHEM-2.3 · rank 19 of 82	€0.072	€0.25
Qwen3-Embedding-0.6B	Embeddings	0.59 / 0.6	from 5 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Embedding-0.6B	MTEB TRECCOVID · rank 2 of 37	€0.011	–
Qwen3-Reranker-4B	Language models	4	from 7 GB	Apache 2.0	huggingface.co/Qwen/Qwen3-Reranker-4B	–	€0.05	–
Qwen3-VL 30B-A3B Instruct	Image models	31.07 / 3	from 22 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct	–	€0.13	€0.54
Qwen3-VL 32B Instruct	Image models	33.36	from 23 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-32B-Instruct	–	–	–
Qwen3-VL 32B Thinking	Image models	33.36	from 23 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-32B-Thinking	–	–	–
Qwen3-VL 4B Instruct	Image models	4.44	from 7 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-4B-Instruct	–	–	–
Qwen3-VL 8B Instruct	Image models	8.77	from 9 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-8B-Instruct	–	€0.22	€0.67
Qwen3-VL 8B Thinking	Image models	8.77	from 9 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-8B-Thinking	–	–	–
Qwen3-VL Embedding 8B	Embeddings	8.14	from 9 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-Embedding-8B	–	€0.08	–
Qwen3-VL Reranker-8B	Language models	8	from 9 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-Reranker-8B	–	€0.04	–
Qwen3-VL-Embedding-2B	Embeddings	2.13	from 6 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-Embedding-2B	–	–	–
Qwen3.5 2B	Language models	2.27	from 6 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.5-2B	–	–	–
Qwen3.5-27B	Language models	27.78	from 20 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.5-27B	MMLU-Pro (TIGER-Lab Leaderboard) · rank 5 of 42	€0.23	€2.33
Qwen3.5-35B-A3B	Language models	35.95 / 3	from 24 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.5-35B-A3B	searchit Entity Extraction · rank 28 of 53	€0.13	€0.90
Qwen3.5-9B	Language models	9.65	from 10 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.5-9B	MMLU-Pro (TIGER-Lab Leaderboard) · rank 8 of 42	€0.09	€0.13
Qwen3.6-27B	Language models	27.78	from 20 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.6-27B	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 54 of 153	€0.22	€1.79
Qwen3.6-35B-A3B	Language models	35.95 / 3	from 24 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.6-35B-A3B	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 64 of 153	€0.13	€0.45
Qwen3.8-27B	Language models	27.78	from 20 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.8-27B	searchit Entity Extraction · rank 27 of 53	€0.09	€0.27
Safety GPT OSS 20B	Language models	21 / 3.6	from 16 GB	Apache-2.0	huggingface.co/openai/gpt-oss-safeguard-20b	–	€0.027	€0.09
SauerkrautLM-ColQwen3-8b-v0.1	Embeddings	8.14	from 9 GB	Apache-2.0	huggingface.co/VAGOsolutions/SauerkrautLM-ColQwen3-8b-v0.1	–	–	–
Seed-OSS-36B-Instruct	Language models	36.15	from 24 GB	Apache-2.0	huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 90 of 153	–	–
siglip-so400m-patch14-384	Embeddings	0.88	from 5 GB	Apache-2.0	huggingface.co/google/siglip-so400m-patch14-384	–	–	–
SNOWFLAKE ARCTIC-EMBED-M-V1.5	Embeddings	0.11	from 5 GB	Apache-2.0	huggingface.co/Snowflake/snowflake-arctic-embed-m-v1.5	MTEB TRECCOVID · rank 6 of 37	€0.022	–
snowflake-arctic-embed-l	Embeddings	0.34	from 5 GB	Apache-2.0	huggingface.co/Snowflake/snowflake-arctic-embed-l	MTEB SciFact · rank 6 of 28	–	–
snowflake-arctic-embed-l-v2.0	Embeddings	0.57	from 5 GB	Apache-2.0	huggingface.co/Snowflake/snowflake-arctic-embed-l-v2.0	MTEB GermanQuAD-Retrieval · rank 2 of 26	–	–
snowflake-arctic-embed-m-long	Embeddings	0.14	from 5 GB	Apache-2.0	huggingface.co/Snowflake/snowflake-arctic-embed-m-long	MTEB ArguAna · rank 11 of 38	–	–
snowflake-arctic-embed-m-v2.0	Embeddings	0.3	from 5 GB	Apache-2.0	huggingface.co/Snowflake/snowflake-arctic-embed-m-v2.0	MTEB LegalQuAD · rank 10 of 44	€0.036	–
text2vec-base-chinese	Embeddings	0.1	from 5 GB	Apache-2.0	huggingface.co/shibing624/text2vec-base-chinese	MTEB LegalQuAD · rank 44 of 44	€0.0045	–
UI-TARS 1.5 7B	Language models	8.29	from 9 GB	Apache-2.0	huggingface.co/ByteDance-Seed/UI-TARS-1.5-7B	–	€0.09	€0.18
Venice SD35	Image generation	8.15	from 9 GB	Stability AI Community License	huggingface.co/stabilityai/stable-diffusion-3.5-large	–	–	–
Venice Uncensored 1.2	Language models	24.01	from 18 GB	Apache-2.0	huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition	–	€0.18	€0.81
Voxtral Mini 1.0	Language models	4.68	from 7 GB	Apache-2.0	huggingface.co/mistralai/Voxtral-Mini-3B-2507	–	€0.018	€0.018
Voxtral Small 24B	Language models	24.26	from 18 GB	Apache-2.0	huggingface.co/mistralai/Voxtral-Small-24B-2507	–	€0.045	€0.13
Voxtral-Mini-3B-2507	Language models	4	from 7 GB	Apache-2.0	huggingface.co/mistralai/Voxtral-Mini-3B-2507	–	–	–
VultronRetrieverPrime-Qwen3.5-8B	Embeddings	8.39	from 9 GB	Apache-2.0	huggingface.co/vultr/VultronRetrieverPrime-Qwen3.5-8B	–	–	–
webAI-ColVec1.1-8b	Embeddings	8.39	from 9 GB	eigene Lizenz - vor Einsatz pruefen	huggingface.co/webAI-Official/webAI-ColVec1.1-8b	–	–	–
Whisper Large V3 Turbo	Language models	0.81 / 0.81	from 5 GB	MIT	huggingface.co/openai/whisper-large-v3-turbo	–	€9.21	€9.21
Whisper large-v3 New	Language models	1.54 / 1.55	from 5 GB	Apache-2.0	huggingface.co/openai/whisper-large-v3	–	€14.61	€14.61
Z-Image Turbo	Image generation	6.16	from 8 GB	Apache-2.0	huggingface.co/Tongyi-MAI/Z-Image-Turbo	–	–	–

API price: cheapest offer worldwide in € per 1M tokens – to compare renting with self-hosting. Best benchmark: the model's best rank in the leaderboards. Source: providers' pricing pages, read and checked by searchit. As of 7 Nov 2026.

48 GB – one RTX PRO 5000

GPU memory (4-bit, derived) above 24 GB up to 48 GB · 13 models

Model	Type	Parameters (bn, total / active)	GPU memory (4-bit, derived)GPU memory	Licence	Model files	Best benchmark	API price: InputAPI price in	API price: OutputAPI price out
Hunyuan Image 3.0	Image generation	80 / 13	from 48 GB	Tencent Hunyuan Community License (Image 3.0)	huggingface.co/tencent/HunyuanImage-3.0	–	–	–
Hunyuan-A13B-Instruct	Language models	80 / 13	from 48 GB	Tencent Hunyuan Community License (A13B)	huggingface.co/tencent/Hunyuan-A13B-Instruct	–	€0.13	€0.51
Llama 3 70B	Language models	70.55	from 43 GB	Llama 3 Community License	huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 137 of 153	€2.37	€3.13
Llama 3.1 70B	Language models	70.55	from 43 GB	Llama 3.1 Community License	huggingface.co/meta-llama/Llama-3.1-70B-Instruct	MMLU-Pro (TIGER-Lab Leaderboard) · rank 27 of 42	€0.31	€1.25
Llama 3.3 70B	Language models	70.6	from 43 GB	Llama 3.3 Community License	huggingface.co/meta-llama/Llama-3.3-70B-Instruct	Vectara HHEM-2.3 · rank 4 of 82	€0.09	€0.29
Macaron V1 Tall	Language models	50	from 32 GB	MIT	huggingface.co/mindlab-research/Macaron-V1-Tall	–	€0.22	€1.28
Mixtral 8x7B Instruct	Language models	46.7 / 13	from 30 GB	Apache-2.0	huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1	MMLU-Pro (TIGER-Lab Leaderboard) · rank 37 of 42	€0.40	€0.63
Qwen2.5-72B-Instruct	Language models	72.7	from 44 GB	Qwen License	huggingface.co/Qwen/Qwen2.5-72B-Instruct	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 123 of 153	€0.32	€0.36
Qwen2.5-VL 72B Instruct	Image models	73.41	from 45 GB	Qwen-Lizenz (nicht Apache)	huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct	–	€0.72	€0.90
Qwen3-Coder-Next (80B)	Language models	80 / 3	from 48 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Coder-Next	–	€0.057	€0.21
Qwen3-Next-80B-Thinking	Language models	80 / 3	from 48 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking	–	€0.13	€1.07
Skyfall 36B v2 (FP8)	Language models	36.91	from 25 GB	–	huggingface.co/TheDrummer/Skyfall-36B-v2	–	€0.49	€0.72
SwissAI Apertus 1.5 70B	Language models	70	from 43 GB	Apache-2.0	huggingface.co/swiss-ai/Apertus-v1.5-70B	–	€0.70	€2.80

96 GB – one RTX PRO 6000

GPU memory (4-bit, derived) above 48 GB up to 96 GB · 20 models

Model	Type	Parameters (bn, total / active)	GPU memory (4-bit, derived)GPU memory	Licence	Model files	Best benchmark	API price: InputAPI price in	API price: OutputAPI price out
@cf/nvidia/nemotron-3-120b-a12b	Language models	123.61 / 12	from 72 GB	NVIDIA Nemotron Open Model License	huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16	–	€0.45	€1.34
Command R+ 04-2024	Language models	103.81	from 62 GB	CC-BY-NC-4.0	huggingface.co/CohereLabs/c4ai-command-r-plus	–	€2.69	€13.43
Devstral 2 123B	Language models	125.03	from 73 GB	Modified MIT	huggingface.co/mistralai/Devstral-2-123B-Instruct-2512	–	€0.43	€2.15
GLM 4.5 Air	Language models	106 / 12	from 63 GB	MIT	huggingface.co/zai-org/GLM-4.5-Air	–	€0.12	€0.76
GLM 4.5V	Image models	106 / 12	from 63 GB	MIT	huggingface.co/zai-org/GLM-4.5V	–	€0.54	€1.61
GLM 4.6V	Image models	106	from 63 GB	MIT	huggingface.co/zai-org/GLM-4.6V	–	€0.27	€0.81
GPT OSS Safeguard 120B	Language models	117 / 5.1	from 69 GB	Apache-2.0	huggingface.co/openai/gpt-oss-safeguard-120b	–	€0.063	€0.27
gpt-oss-120b	Language models	117 / 5.1	from 69 GB	Apache-2.0	huggingface.co/openai/gpt-oss-120b	MMLU-Pro (TIGER-Lab Leaderboard) · rank 10 of 42	€0.026	€0.12
Ling 3.0 Flash	Language models	124 / 5.1	from 73 GB	MIT	huggingface.co/inclusionAI/Ling-3.0-flash	–	€0.054	€0.16
Ling 3.0 Flash Fin	Language models	124 / 5.1	from 73 GB	MIT	huggingface.co/inclusionAI/Ling-3.0-flash-Fin	–	€0.067	€0.20
Ling 3.0 Flash VL	Language models	124 / 5.5	from 73 GB	MIT	huggingface.co/inclusionAI/Ling-3.0-flash-VL	–	€0.067	€0.20
Llama 3.2 90B	Language models	88.59	from 53 GB	Llama 3.2 Community License	huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 141 of 153	€0.64	€0.64
Llama 4 Scout 17B-16E	Image models	108.64 / 17	from 64 GB	Llama 4 Community License	huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct	Vectara HHEM-2.3 · rank 29 of 82	€0.09	€0.27
Mistral Large 2.1	Language models	123	from 72 GB	proprietär	huggingface.co/mistralai/Mistral-Large-Instruct-2411	Vectara HHEM-2.3 · rank 6 of 82	–	–
Mistral Medium 3.5	Language models	127.7	from 75 GB	Modified MIT (Umsatzgrenze 20 Mio. USD/Monat)	huggingface.co/mistralai/Mistral-Medium-3.5-128B	searchit Entity Extraction · rank 9 of 53	€1.50	€6.00
Mistral Small	Language models	119.4 / 6.5	from 70 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Small-4-119B-2603	–	€0.90	€2.69
NVIDIA Nemotron 3 Super 120B A12B	Language models	120 / 12	from 70 GB	NVIDIA Nemotron Open Model License	huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16	searchit Entity Extraction · rank 38 of 53	€0.067	€0.29
Pixtral Large 25.02	Language models	124	from 73 GB	Mistral Research License (MRL)	huggingface.co/mistralai/Pixtral-Large-Instruct-2411	–	€1.79	€5.37
Qwen 3 Next 80b	Language models	81.33 / 3	from 49 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct	searchit Entity Extraction · rank 24 of 53	€0.063	€0.54
Qwen3.5-122B-A10B	Image models	125.09 / 10	from 73 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.5-122B-A10B	Vectara HHEM-2.3 · rank 54 of 82	€0.22	€1.57

192 GB – two RTX PRO 6000

GPU memory (4-bit, derived) above 96 GB up to 192 GB · 18 models

Model	Type	Parameters (bn, total / active)	GPU memory (4-bit, derived)GPU memory	Licence	Model files	Best benchmark	API price: InputAPI price in	API price: OutputAPI price out
DeepSeek V4 Flash	Language models	290.94 / 13	from 165 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-V4-Flash	searchit Entity Extraction · rank 21 of 53	€0.081	€0.16
DeepSeek V4 Flash 0423	Language models	290.94 / 13	from 165 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-V4-Flash	–	€0.12	€0.25
DeepSeek V4 Flash 0731	Language models	304.18	from 172 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 40 of 153	€0.054	€0.16
deepseek-v4-flash-vision-exp	Language models	304.65	from 172 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp	–	€0.39	€1.18
GLM 5.3 Flash	Language models	320 / 18	from 180 GB	MIT	huggingface.co/zai-org/GLM-5.3-Flash	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 43 of 153	€0.09	€0.31
Hy3	Language models	295 / 21	from 167 GB	Apache-2.0	huggingface.co/tencent/Hy3	–	€0.12	€0.47
MiMo v2.5	Language models	310 / 15	from 175 GB	MIT	huggingface.co/XiaomiMiMo/MiMo-V2.5	searchit Entity Extraction · rank 11 of 53	€0.13	€0.25
MiMo V2.6 Flash	Language models	309 / 15	from 174 GB	MIT	huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL	–	€0.13	€0.26
Minimax M2	Language models	230 / 10	from 131 GB	Modified-MIT	huggingface.co/MiniMaxAI/MiniMax-M2	–	€0.13	€0.54
Minimax M2.1	Language models	229 / 10	from 130 GB	Modified-MIT	huggingface.co/MiniMaxAI/MiniMax-M2.1	Vectara HHEM-2.3 · rank 56 of 82	€0.13	€0.54
MiniMax M2.5	Language models	228.7 / 10	from 130 GB	MiniMax Model License	huggingface.co/MiniMaxAI/MiniMax-M2.5	MMLU-Pro (TIGER-Lab Leaderboard) · rank 11 of 42	€0.13	€0.54
MiniMax M2.7	Language models	228.69 / 10	from 130 GB	MiniMax Non-Commercial License	huggingface.co/MiniMaxAI/MiniMax-M2.7	Vectara HHEM-2.3 · rank 67 of 82	€0.27	€1.07
Qwen 3 235B A22B FP8	Language models	235 / 22	from 134 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-235B-A22B-FP8	–	€0.18	€0.54
Qwen 3 235B A22B Thinking 2507	Language models	235.09 / 22	from 134 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 61 of 153	€0.40	€3.13
Qwen3 235B A22B Instruct 2507	Language models	235.09 / 22	from 134 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507	searchit Entity Extraction · rank 17 of 53	€0.099	€0.39
Qwen3-VL 235B-A22B Instruct	Image models	235.67 / 22	from 134 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct	–	€0.18	€0.79
Qwen3.8-Flash	Language models	180 / 6	from 103 GB	proprietär	huggingface.co/Qwen/Qwen3.8-Flash-Next	searchit Entity Extraction · rank 36 of 53	€0.13	€0.44
Step 3.7 Flash	Language models	198 / 11	from 113 GB	Apache-2.0	huggingface.co/stepfun-ai/Step-3.7-Flash	–	€0.18	€1.03

384 GB – four RTX PRO 6000

GPU memory (4-bit, derived) above 192 GB up to 384 GB · 19 models

Model	Type	Parameters (bn, total / active)	GPU memory (4-bit, derived)GPU memory	Licence	Model files	Best benchmark	API price: InputAPI price in	API price: OutputAPI price out
DeepSeek R1 Turbo	Language models	684.49 / 37	from 381 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-R1	–	–	–
DeepSeek V3.1	Language models	671 / 37	from 374 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-V3.1	Vectara HHEM-2.3 · rank 14 of 82	€0.22	€0.85
DeepSeek-R1 (0528)	Language models	684.53 / 37	from 381 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-R1-0528	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 93 of 153	€0.45	€1.93
DeepSeek-V3	Language models	684.53 / 37	from 381 GB	DeepSeek Model License	huggingface.co/deepseek-ai/DeepSeek-V3	Vectara HHEM-2.3 · rank 20 of 82	€0.29	€0.80
deepseek-v3.2	Language models	685.36	from 381 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-V3.2	Vectara HHEM-2.3 · rank 22 of 82	€0.23	€0.34
GLM 4.6	Language models	357	from 201 GB	MIT	huggingface.co/zai-org/GLM-4.6	Berkeley Function-Calling Leaderboard V4 · rank 3 of 42	€0.39	€1.57
GLM 4.7	Language models	358	from 201 GB	MIT	huggingface.co/zai-org/GLM-4.7	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 73 of 153	€0.27	€0.99
GLM-4.5	Language models	355 / 32	from 200 GB	MIT	huggingface.co/zai-org/GLM-4.5	–	€0.54	€1.97
H Hermes 4 405B	Language models	405.85	from 228 GB	–	huggingface.co/NousResearch/Hermes-4-405B	–	€0.89	€2.68
Llama 4 Maverick	Language models	401.58 / 17	from 225 GB	Llama 4 Community License	huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct	Vectara HHEM-2.3 · rank 31 of 82	€0.16	€0.52
MiniMax M1	Language models	456 / 45.9	from 255 GB	Apache-2.0	huggingface.co/MiniMaxAI/MiniMax-M1-80k	–	€0.49	€1.97
MiniMax M3	Language models	427.04 / 23	from 239 GB	eigene Lizenz	huggingface.co/MiniMaxAI/Minimax-M3	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 84 of 153	€0.27	€1.07
Mistral Large	Language models	675 / 41	from 376 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512	–	€0.45	€1.34
Mistral Large 3	Language models	675 / 41	from 376 GB	Apache-2.0	huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512	Vectara HHEM-2.3 · rank 72 of 82	€0.22	€0.67
NVIDIA Nemotron 3 Ultra	Language models	550 / 55	from 307 GB	OpenMDW-1.1	huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 61 of 153	€0.45	€1.97
Qwen 3 Coder 480B Turbo	Language models	480 / 35	from 268 GB	Apache-2.0	huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct	–	€0.099	€0.81
Qwen3.5-397B-A17B	Language models	403.4 / 17	from 226 GB	Apache-2.0	huggingface.co/Qwen/Qwen3.5-397B-A17B	MMLU-Pro (TIGER-Lab Leaderboard) · rank 2 of 42	€0.40	€2.69
R1	Language models	671 / 37	from 374 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-R1	–	€1.21	€4.84
Trinity Large (Thinking)	Language models	398 / 13	from 223 GB	OpenMDW-1.1	huggingface.co/arcee-ai/Trinity-Large-Thinking	–	€0.20	€0.76

768 GB – eight RTX PRO 6000

GPU memory (4-bit, derived) above 384 GB up to 768 GB · 14 models

Model	Type	Parameters (bn, total / active)	GPU memory (4-bit, derived)GPU memory	Licence	Model files	Best benchmark	API price: InputAPI price in	API price: OutputAPI price out
GLM 5.1	Language models	754	from 419 GB	–	huggingface.co/zai-org/GLM-5.1	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 45 of 153	€0.81	€2.60
glm-5.2	Language models	753	from 419 GB	MIT	huggingface.co/zai-org/GLM-5.2	SWE-bench Verified (Epoch AI) · rank 3 of 33	€0.15	€0.48
glm-5.3	Language models	753.3	from 419 GB	GLM-5.3 License	huggingface.co/zai-org/GLM-5.3	FrontierMath Tiers 1-3 v2 (Epoch AI) · rank 30 of 91	€0.18	€0.57
Hy4 Preview	Language models	770 / 49	from 428 GB	Apache-2.0	huggingface.co/tencent/Hy4-preview	–	€0.75	€2.24
Inkling	Language models	975 / 41	from 541 GB	Apache-2.0	huggingface.co/thinkingmachines/Inkling	–	–	–
Kimi K2 Instruct	Language models	1,000 / 32	from 554 GB	Modified MIT	huggingface.co/moonshotai/Kimi-K2-Instruct	Berkeley Function-Calling Leaderboard V4 · rank 8 of 42	€0.51	€2.06
Kimi K2 Instruct 0905	Language models	1,000 / 32	from 554 GB	Modified MIT	huggingface.co/moonshotai/Kimi-K2-Instruct-0905	Vectara HHEM-2.3 · rank 76 of 82	€0.51	€2.06
Kimi K2 Thinking	Language models	1,000 / 32	from 554 GB	Modified MIT	huggingface.co/moonshotai/Kimi-K2-Thinking	–	€0.27	€1.12
Kimi K2.5 (Retiring)	Language models	1,000 / 32	from 554 GB	Modified MIT	huggingface.co/moonshotai/Kimi-K2.5	searchit Entity Extraction · rank 8 of 53	€0.27	€1.34
Kimi K2.6	Language models	1,026.88 / 32	from 569 GB	eigene Lizenz	huggingface.co/moonshotai/Kimi-K2.6	searchit Entity Extraction · rank 10 of 53	€0.35	€1.61
Kimi K2.7 Code	Language models	1,000 / 32	from 554 GB	–	huggingface.co/moonshotai/Kimi-K2.7-Code	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 33 of 153	€0.59	€2.96
Macaron V1 Venti	Language models	748	from 416 GB	MIT	huggingface.co/mindlab-research/Macaron-V1-Venti	–	€0.74	€2.22
MiMo V2.6 Pro	Language models	1,020 / 42	from 565 GB	MIT	huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL	–	€0.78	€1.56
Mistral Large 4	Language models	1,050 / 52	from 582 GB	–	–	searchit Entity Extraction · rank 3 of 53	€0.50	€1.75

Data centre – more than eight cards

GPU memory (4-bit, derived) above 768 GB · 5 models

Model	Type	Parameters (bn, total / active)	GPU memory (4-bit, derived)GPU memory	Licence	Model files	Best benchmark	API price: InputAPI price in	API price: OutputAPI price out
DeepSeek V4 Pro	Language models	1,598.84 / 49	from 884 GB	MIT	huggingface.co/deepseek-ai/DeepSeek-V4-Pro	searchit Entity Extraction · rank 6 of 53	€0.90	€1.79
DeepSeek V4 Pro 0813	Language models	1,650.5 / 49	from 912 GB	–	huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 22 of 153	€1.16	€2.33
Kimi K3	Language models	2,779.93 / 104	from 1,533 GB	eigene Lizenz	huggingface.co/moonshotai/Kimi-K3	searchit Entity Extraction · rank 7 of 53	€1.66	€8.51
Qwen 3.8 2.4T	Language models	2,400 / 95	from 1,324 GB	Qwen3.8-Max License	huggingface.co/Qwen/Qwen3.8-2.4T-A95B	–	€2.24	€6.72
Qwen3.8-Max	Language models	2,446.18 / 95	from 1,350 GB	Qwen3.8-Max License	huggingface.co/Qwen/Qwen3.8-2.4T-A95B	OTIS Mock AIME 2024-2025 (Epoch AI) · rank 13 of 153	€1.48	€4.43

23 further models with open weights have no GPU memory figure; they are in the model list.