Datasets:
id stringlengths 5 24 | name stringlengths 9 32 | maker stringlengths 4 49 | params_total int64 1.73B 1,049B | q4_k_m_bytes float64 1.1B 635B ⌀ | q8_0_bytes float64 1.85B 1,115B ⌀ | bf16_bytes float64 3.47B 2,099B ⌀ | kv_bytes_per_token_fp16 float64 890 328k ⌀ | context_max int64 32.8k 1.05M | quant_note stringclasses 8
values | page stringlengths 36 55 | checked stringdate 2026-10-23 00:00:00 2026-10-31 00:00:00 | q4_k_m_gb float64 1.1 635 ⌀ | q8_0_gb float64 1.85 1.12k ⌀ | bf16_gb float64 3.47 2.1k ⌀ | kv_mb_per_1k_tokens float64 0.9 328 ⌀ | sources stringlengths 138 268 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
llama-3-2-3b | Llama 3.2 3B Instruct | Meta | 3,212,749,824 | 2,019,377,696 | 3,421,899,296 | 6,433,687,840 | 114,688 | 131,072 | null | https://vramneeded.com/models/llama-3-2-3b/ | 2026-10-23 | 2.02 | 3.42 | 6.43 | 114.7 | https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF/tree/main https://huggingface.co/unsloth/Llama-3.2-3B-Instruct/blob/main/config.json https://ollama.com/library/llama3.2 |
llama-3-1-8b | Llama 3.1 8B Instruct | Meta | 8,030,261,248 | 4,920,739,232 | 8,540,775,840 | 16,060,522,496 | 131,072 | 131,072 | null | https://vramneeded.com/models/llama-3-1-8b/ | 2026-10-23 | 4.92 | 8.54 | 16.06 | 131.1 | https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF/tree/main https://huggingface.co/unsloth/Meta-Llama-3.1-8B-Instruct/blob/main/config.json https://ollama.com/library/llama3.1 |
deepseek-r1-8b | DeepSeek-R1-0528-Qwen3-8B | DeepSeek | 8,190,735,360 | 5,027,783,040 | 8,709,517,696 | 16,388,042,816 | 147,456 | 131,072 | null | https://vramneeded.com/models/deepseek-r1-8b/ | 2026-10-23 | 5.03 | 8.71 | 16.39 | 147.5 | https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B https://huggingface.co/bartowski/deepseek-ai_DeepSeek-R1-0528-Qwen3-8B-GGUF/tree/main https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/main/config.json https://ollama.com/library/deepseek-r1 |
gpt-oss-20b | gpt-oss-20b | OpenAI | 20,914,757,184 | 11,673,418,816 | 12,109,566,016 | 13,792,637,760 | 24,576 | 131,072 | The MoE experts stay in MXFP4 in every GGUF, so Q4_K_M, Q8_0 and BF16 files are close in size. | https://vramneeded.com/models/gpt-oss-20b/ | 2026-10-23 | 11.67 | 12.11 | 13.79 | 24.6 | https://huggingface.co/openai/gpt-oss-20b https://huggingface.co/bartowski/openai_gpt-oss-20b-GGUF/tree/main https://huggingface.co/openai/gpt-oss-20b/blob/main/config.json https://ollama.com/library/gpt-oss |
qwen3-14b | Qwen3 14B | Alibaba Qwen | 14,768,307,200 | 9,001,753,632 | 15,698,534,432 | 29,543,423,776 | 163,840 | 131,072 | null | https://vramneeded.com/models/qwen3-14b/ | 2026-10-31 | 9 | 15.7 | 29.54 | 163.8 | https://huggingface.co/Qwen/Qwen3-14B https://huggingface.co/bartowski/Qwen_Qwen3-14B-GGUF/tree/main https://huggingface.co/Qwen/Qwen3-14B/blob/main/config.json https://ollama.com/library/qwen3/tags |
gemma-4-26b-a4b | Gemma 4 26B A4B | Google | 25,805,936,206 | 17,035,039,872 | 26,859,861,120 | 50,505,135,424 | null | 131,072 | null | https://vramneeded.com/models/gemma-4-26b-a4b/ | 2026-10-23 | 17.04 | 26.86 | 50.51 | null | https://huggingface.co/google/gemma-4-26B-A4B-it https://huggingface.co/bartowski/google_gemma-4-26B-A4B-it-GGUF/tree/main https://huggingface.co/google/gemma-4-26B-A4B-it/blob/main/config.json https://ollama.com/library/gemma4 |
qwen3-8-27b | Qwen3.8 27B | Alibaba Qwen | 27,781,427,952 | 17,442,399,968 | 29,116,388,960 | 54,657,734,144 | 65,536 | 262,144 | null | https://vramneeded.com/models/qwen3-8-27b/ | 2026-10-31 | 17.44 | 29.12 | 54.66 | 65.5 | https://huggingface.co/Qwen/Qwen3.8-27B https://huggingface.co/bartowski/Qwen3.8-27B-GGUF/tree/main https://huggingface.co/Qwen/Qwen3.8-27B/blob/main/config.json https://ollama.com/library/qwen3.8 |
qwen3-8-35b-a3b | Qwen3.8-35B-A3B (Empero Distill) | Empero | 35,107,181,936 | 21,713,462,944 | 37,802,149,536 | 71,066,994,688 | 20,480 | 262,144 | GGUF sizes from Empero's own Distill-GGUF repo (imatrix-calibrated IQ* and K-quants). | https://vramneeded.com/models/qwen3-8-35b-a3b/ | 2026-10-28 | 21.71 | 37.8 | 71.07 | 20.5 | https://huggingface.co/empero-ai/Qwen3.8-35B-A3B-Distill https://huggingface.co/empero-ai/Qwen3.8-35B-A3B-Distill-GGUF/tree/main https://huggingface.co/empero-ai/Qwen3.8-35B-A3B-Distill/blob/main/config.json |
gemma-4-31b | Gemma 4 31B | Google | 31,273,088,876 | 19,598,489,952 | 32,635,677,024 | 61,413,189,920 | null | 131,072 | null | https://vramneeded.com/models/gemma-4-31b/ | 2026-10-23 | 19.6 | 32.64 | 61.41 | null | https://huggingface.co/google/gemma-4-31B-it https://huggingface.co/bartowski/google_gemma-4-31B-it-GGUF/tree/main https://huggingface.co/google/gemma-4-31B-it/blob/main/config.json https://ollama.com/library/gemma4 |
llama-3-1-70b | Llama 3.1 70B Instruct | Meta | 70,553,706,496 | 42,520,398,400 | 74,975,054,592 | 141,107,412,992 | 327,680 | 131,072 | null | https://vramneeded.com/models/llama-3-1-70b/ | 2026-10-31 | 42.52 | 74.98 | 141.11 | 327.7 | https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct https://huggingface.co/bartowski/Meta-Llama-3.1-70B-Instruct-GGUF/tree/main https://huggingface.co/unsloth/Meta-Llama-3.1-70B-Instruct/blob/main/config.json https://ollama.com/library/llama3.1 |
gpt-oss-120b | gpt-oss-120b | OpenAI | 116,829,156,672 | 62,841,713,344 | 63,387,346,368 | 65,369,016,768 | 36,864 | 131,072 | Experts stay in MXFP4, so quant sizes are close. | https://vramneeded.com/models/gpt-oss-120b/ | 2026-10-23 | 62.84 | 63.39 | 65.37 | 36.9 | https://huggingface.co/openai/gpt-oss-120b https://huggingface.co/bartowski/openai_gpt-oss-120b-GGUF/tree/main https://huggingface.co/openai/gpt-oss-120b/blob/main/config.json https://ollama.com/library/gpt-oss |
qwen3-8-flash-next | Qwen3.8 Flash Next | Alibaba Qwen | 179,999,981,459 | 119,599,030,080 | 188,260,769,536 | 354,029,929,568 | 24,576 | 262,144 | null | https://vramneeded.com/models/qwen3-8-flash-next/ | 2026-10-27 | 119.6 | 188.26 | 354.03 | 24.6 | https://huggingface.co/Qwen/Qwen3.8-Flash-Next https://huggingface.co/bartowski/Qwen3.8-Flash-Next-GGUF/tree/main https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/main/config.json |
deepseek-v4-1-flash-1017 | DeepSeek-V4.1-Flash-1017 | DeepSeek | 763,205,315,794 | 500,707,576,573 | 508,376,763,197 | null | 890 | 1,048,576 | No Q4_K_M or Q8_0 build exists; 4-bit and 8-bit columns are Unsloth UD-Q4_K_XL and UD-Q8_K_XL (experts stay MXFP4). With the Engram table left on SSD, UD-Q4_K_XL needs 297.9 GB of RAM/VRAM and the 1-bit UD-IQ1_S needs 122.4 GB. | https://vramneeded.com/models/deepseek-v4-1-flash-1017/ | 2026-10-31 | 500.71 | 508.38 | null | 0.9 | https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017 https://huggingface.co/unsloth/DeepSeek-V4.1-Flash-1017-GGUF/tree/main https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash https://unsloth.ai/docs/models/deepseek-v4.1-flash |
mistral-large-4 | Mistral Large 4 (Le Chonk) | Mistral AI | 1,049,357,365,536 | 634,802,244,608 | 1,115,150,340,608 | 2,098,861,021,920 | 83,840 | 1,048,576 | null | https://vramneeded.com/models/mistral-large-4/ | 2026-10-31 | 634.8 | 1,115.15 | 2,098.86 | 83.8 | https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B https://huggingface.co/ggml-org/Mistral-Large-4-1T-A52B-GGUF/tree/main https://mistral.ai/news/mistral-large-4-oss/ |
flint | Recursive Flint | Recursive | 39,785,287,456 | 25,054,688,704 | 42,301,076,928 | 79,576,479,200 | 28,672 | 262,144 | No Q4_K_M build in the Unsloth repo; the 4-bit column is Unsloth UD-Q4_K_XL. Ollama's default flint tag is 23 GB. | https://vramneeded.com/models/flint/ | 2026-10-24 | 25.05 | 42.3 | 79.58 | 28.7 | https://huggingface.co/RecursiveSI/flint https://huggingface.co/unsloth/flint-GGUF/tree/main https://huggingface.co/RecursiveSI/flint/blob/main/config.json https://ollama.com/library/flint/tags |
ornith-1-5-122b | Ornith-1.5-122B | Ornith | 125,086,497,008 | 79,713,771,665 | 132,555,710,737 | 249,362,981,521 | 24,576 | 262,144 | No Q4_K_M build in the Unsloth repo; the 4-bit column is Unsloth UD-Q4_K_XL (3 files). Ollama's ornith-1.5:122b-q4_K_M tag is 77 GB. | https://vramneeded.com/models/ornith-1-5-122b/ | 2026-10-25 | 79.71 | 132.56 | 249.36 | 24.6 | https://huggingface.co/ornith-ai/Ornith-1.5-122B https://huggingface.co/unsloth/Ornith-1.5-122B-GGUF/tree/main https://huggingface.co/ornith-ai/Ornith-1.5-122B/blob/main/config.json https://ollama.com/library/ornith-1.5/tags |
nanbeige4-3-33b | Nanbeige4.3 33B | Nanbeige | 33,118,977,024 | 20,788,825,216 | 35,195,128,960 | 66,243,464,192 | 212,992 | 524,288 | null | https://vramneeded.com/models/nanbeige4-3-33b/ | 2026-10-26 | 20.79 | 35.2 | 66.24 | 213 | https://huggingface.co/Nanbeige/Nanbeige4.3-33B https://huggingface.co/bartowski/Nanbeige_Nanbeige4.3-33B-GGUF/tree/main https://huggingface.co/Nanbeige/Nanbeige4.3-33B/blob/main/config.json |
lfm2-5-24b-a2b | LFM2.5-24B-A2B | Liquid AI | 23,971,585,280 | 14,526,258,464 | 25,493,732,640 | 47,962,094,880 | 20,480 | 131,072 | null | https://vramneeded.com/models/lfm2-5-24b-a2b/ | 2026-10-27 | 14.53 | 25.49 | 47.96 | 20.5 | https://huggingface.co/LiquidAI/LFM2.5-24B-A2B https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-GGUF/tree/main https://huggingface.co/LiquidAI/LFM2.5-24B-A2B/blob/main/config.json |
lfm3-1-7b | LFM3 1.7B | Liquid AI | 1,728,801,536 | 1,099,627,808 | 1,846,238,496 | 3,466,626,336 | 12,288 | 262,144 | null | https://vramneeded.com/models/lfm3-1-7b/ | 2026-10-31 | 1.1 | 1.85 | 3.47 | 12.3 | https://huggingface.co/LiquidAI/LFM3-1.7B https://huggingface.co/LiquidAI/LFM3-1.7B-GGUF/tree/main https://huggingface.co/LiquidAI/LFM3-1.7B/blob/main/config.json |
spark-x2-6-4b | Spark-X2.6 4B | XHToken | 4,112,079,360 | 3,543,441,312 | 4,375,021,824 | 8,229,921,024 | 36,864 | 1,048,576 | null | https://vramneeded.com/models/spark-x2-6-4b/ | 2026-10-31 | 3.54 | 4.38 | 8.23 | 36.9 | https://huggingface.co/XHToken/Spark-X2.6-4B https://huggingface.co/XHToken/Spark-X2.6-4B-GGUF/tree/main https://huggingface.co/XHToken/Spark-X2.6-4B/blob/main/config.json |
qwen3-6-35b-a3b | Qwen3.6 35B-A3B | Alibaba Qwen | 35,951,822,704 | 22,285,080,192 | 37,812,647,552 | 71,065,941,536 | 20,480 | 262,144 | null | https://vramneeded.com/models/qwen3-6-35b-a3b/ | 2026-10-31 | 22.29 | 37.81 | 71.07 | 20.5 | https://huggingface.co/Qwen/Qwen3.6-35B-A3B https://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF/tree/main https://huggingface.co/Qwen/Qwen3.6-35B-A3B/blob/main/config.json |
mistral-7b | Mistral 7B Instruct v0.3 | Mistral AI | 7,248,023,552 | 4,372,812,000 | 7,702,565,088 | 14,496,047,104 | 131,072 | 32,768 | null | https://vramneeded.com/models/mistral-7b/ | 2026-10-31 | 4.37 | 7.7 | 14.5 | 131.1 | https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3 https://huggingface.co/bartowski/Mistral-7B-Instruct-v0.3-GGUF/tree/main https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/blob/main/config.json |
mixtral-8x7b | Mixtral 8x7B Instruct v0.1 | Mistral AI | 46,702,792,704 | 26,441,533,376 | 49,624,262,592 | 93,405,585,408 | 131,072 | 32,768 | null | https://vramneeded.com/models/mixtral-8x7b/ | 2026-10-31 | 26.44 | 49.62 | 93.41 | 131.1 | https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1 https://huggingface.co/TheBloke/Mixtral-8x7B-Instruct-v0.1-GGUF/tree/main https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1/blob/main/config.json https://mistral.ai/news/mixtral-of-experts |
qwen2-5-7b | Qwen2.5 7B Instruct | Alibaba Qwen | 7,615,616,512 | 4,683,074,240 | 8,098,525,888 | 15,237,853,600 | 57,344 | 131,072 | null | https://vramneeded.com/models/qwen2-5-7b/ | 2026-10-31 | 4.68 | 8.1 | 15.24 | 57.3 | https://huggingface.co/Qwen/Qwen2.5-7B-Instruct https://huggingface.co/bartowski/Qwen2.5-7B-Instruct-GGUF/tree/main https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/main/config.json |
qwen2-5-14b | Qwen2.5 14B Instruct | Alibaba Qwen | 14,770,033,664 | 8,988,110,976 | 15,701,598,336 | 29,547,716,448 | 196,608 | 131,072 | null | https://vramneeded.com/models/qwen2-5-14b/ | 2026-10-31 | 8.99 | 15.7 | 29.55 | 196.6 | https://huggingface.co/Qwen/Qwen2.5-14B-Instruct https://huggingface.co/bartowski/Qwen2.5-14B-Instruct-GGUF/tree/main https://huggingface.co/Qwen/Qwen2.5-14B-Instruct/blob/main/config.json |
qwen2-5-32b | Qwen2.5 32B Instruct | Alibaba Qwen | 32,763,876,352 | 19,851,336,576 | 34,820,885,376 | 65,535,970,112 | 262,144 | 131,072 | null | https://vramneeded.com/models/qwen2-5-32b/ | 2026-10-31 | 19.85 | 34.82 | 65.54 | 262.1 | https://huggingface.co/Qwen/Qwen2.5-32B-Instruct https://huggingface.co/bartowski/Qwen2.5-32B-Instruct-GGUF/tree/main https://huggingface.co/Qwen/Qwen2.5-32B-Instruct/blob/main/config.json |
nemotron-3-5-transcribe | Nemotron 3.5 Transcribe 30B-A4B | NVIDIA | 33,697,781,029 | null | 36,675,053,312 | 67,496,013,734 | 6,144 | 65,536 | Speech-to-text model, not a chat LLM. No Q4_K_M build is published; 4-bit-class builds (AD-IQ4_NL 21.3 GB, NVFP4) are on its page. Q8_0 = GGUF decoder + encoder. | https://vramneeded.com/models/nemotron-3-5-transcribe/ | 2026-10-31 | null | 36.68 | 67.5 | 6.1 | https://vramneeded.com/models/nemotron-3-5-transcribe/ https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B https://huggingface.co/AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF/tree/main |
qwen3-8-flash-next-coder | Qwen3.8-Flash-Next Coder | Alibaba Qwen (pruned by ISTA-DASLab / AgentionAI) | 176,900,000,000 | null | null | null | 24,576 | 262,144 | Expert-pruned community builds only (no standard Q4_K_M/Q8_0/BF16 GGUF). Resident weights are 21.4–29.6 GB plus a 28.8 GB+ n-gram table in RAM or on SSD; see its page. | https://vramneeded.com/models/qwen3-8-flash-next-coder/ | 2026-10-31 | null | null | null | 24.6 | https://vramneeded.com/models/qwen3-8-flash-next-coder/ https://huggingface.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF/tree/main |
qwen3-32b | Qwen3 32B | Alibaba Qwen | 32,762,123,264 | 19,762,149,696 | 34,817,719,616 | 65,531,575,584 | 262,144 | 131,072 | null | https://vramneeded.com/models/qwen3-32b/ | 2026-10-31 | 19.76 | 34.82 | 65.53 | 262.1 | https://huggingface.co/Qwen/Qwen3-32B https://huggingface.co/bartowski/Qwen_Qwen3-32B-GGUF/tree/main https://huggingface.co/Qwen/Qwen3-32B/blob/main/config.json |
llama-3-3-70b | Llama 3.3 70B Instruct | Meta | 70,553,706,496 | 42,520,398,816 | 74,975,055,008 | 141,117,918,624 | 327,680 | 131,072 | null | https://vramneeded.com/models/llama-3-3-70b/ | 2026-10-31 | 42.52 | 74.98 | 141.12 | 327.7 | https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct https://huggingface.co/bartowski/Llama-3.3-70B-Instruct-GGUF/tree/main https://huggingface.co/unsloth/Llama-3.3-70B-Instruct/blob/main/config.json |
VRAMNeeded – LLM VRAM requirements
How much GPU/unified memory do open-weight LLMs need? This dataset records real GGUF file sizes (Q4_K_M, Q8_0, BF16 and every other quant we track) and FP16 KV-cache bytes per token for popular local models, each row linked to its Hugging Face source.
Source and interactive version: https://vramneeded.com/compare/ · calculator: https://vramneeded.com/tools/vram-calculator/
Configs
models(30 rows): one row per model, headline quants + KV cache + sources.quant_sizes(514 rows): every quant size for 30 models.
Estimating memory
need ≈ quant file size + kv_bytes_per_token_fp16 × context_tokens (+ runtime overhead).
Hybrid-attention models only count the layers whose cache grows with context.
Limitations
Snapshot dated 2026-10-23 to 2026-10-31. File sizes ≠ exact runtime use (overhead, vision encoders, batch size vary). Blank cells mean no published figure was found.
License & citation
CC BY 4.0. Please credit "VRAMNeeded (vramneeded.com)" with a link.
- Downloads last month
- 34