RunsOnMy ModelsGPUs & MacsBy VRAMBest GPUCan I run…?API pricesVRAM calculatorNew

Local LLM Hardware Index: November 17, 2026

A weekly, automatically generated snapshot of how much memory popular open AI models need and what hardware runs them. Edition 2026-11-17, built from data fetched 2026-11-17 23:34 UTC.

10.5 GBmedian VRAM, top 60 (4-bit, 8K)
48%of top 60 run on 8 GB
72%of top 60 run on 24 GB
17of top 60 need more than 48 GB
2new tracked models this week

Key findings

How much VRAM the top 60 models need

VRAM needed by the 60 most-downloaded open modelsVRAM needed by the 60 most-downloaded open modelsGB at 4-bit (Q4_K_M), 8K context, log scale · estimate from published formulas · edition 2026-11-1712481624481285121024SmolVLM2 500M Video Instruct1.1Qwen2.5 0.5B Instruct1.2Qwen3.5 0.8B1.5Llama 3.2 1B Instruct1.6Llama 3.2 1B1.6Qwen2.5 1.5B Instruct1.9Qwen3 0.6B2HunyuanOCR2Qwen3.5 2B2.3MiniCPM5 2B2.5Qwen3 1.7B2.8Qwen2.5 VL 3B Instruct2.8Unlimited OCR2.9Qwen2.5 3B Instruct2.9DeepSeek OCR2.9Llama 3.2 3B Instruct3.4Gemma 3 4B IT3.5Qwen3.5 4B3.9Qwen3 VL 4B Instruct4.1Qwen3 4B4.1Qwen3 4B Instruct 25074.1Qwen2.5 7B Instruct5.5Qwen2.5 VL 7B Instruct5.5Qwen2.5 Coder 7B Instruct5.5Mistral 7B Instruct v0.25.5Llama 3.1 8B Instruct6.2Qwen3 VL 8B Instruct6.5Qwen3 8B6.5Qwen3.5 9B6.9Qwen3 14B10.3Qwen2.5 Coder 14B Instruct10.6gpt-oss 20B11.8DiffusionGemma 26B A4B IT17.1Gemma 4 26B A4B IT17.3Qwen3.8 27B17.6Qwen3.6 27B18.1Qwen3.5 27B18.1GLM 4.7 Flash18.3Qwen3 Coder 30B A3B Instruct18.8Qwen3 32B21.1Gemma 4 31B IT21.2Qwen3.6 35B A3B21.8Qwen3.5 35B A3B21.8gpt-oss 120B59.6GLM 4.5 Air70.6Qwen3.8 Flash Next112.4MiniMax M2.5131.6DeepSeek V4 Flash167Hy3172.8DeepSeek V4 Flash 0731174.6GLM 5.3 Flash183.7DeepSeek R1377.8DeepSeek V3377.8DeepSeek V3.2378.9GLM 5426.1GLM 5.1429.6GLM 5.3430.4DeepSeek V4.1 Flash435.8Inkling545Kimi K31584.1Dark bars fit a 24 GB card. Source: RunsOnMy Local LLM Hardware Index 2026-11-17, runsonmy.com/report/2026-11-17/ (CC BY 4.0)

SVG · CSV

Share of popular models that run on 8, 12, 16 and 24 GB

Share of the top 60 models that run on 8, 12, 16 and 24 GB cardsShare of the top 60 models that run on 8, 12, 16 and 24 GB cardsTop 60 open models by downloads that run fully in VRAM at 8K context (estimate) · blue: 4-bit · green: 8-bit8 GB48% (29/60)8-bit 21/6012 GB52% (31/60)8-bit 29/6016 GB53% (32/60)8-bit 30/6024 GB72% (43/60)8-bit 32/60Runs = needs at most 95% of VRAM. Source: RunsOnMy Local LLM Hardware Index 2026-11-17 (CC BY 4.0)

SVG · CSV

Cheapest GPU per VRAM tier

Cheapest GPU by VRAM size (launch MSRP)Cheapest GPU by VRAM size (launch MSRP)Cheapest listed card with at least this much VRAM, by US launch MSRP (not street price)8 GB+$249 · Arc B580 (12 GB)12 GB+$249 · Arc B580 (12 GB)16 GB+$349 · Arc A770 16GB (16 GB)24 GB+$999 · RX 7900 XTX (24 GB)32 GB+$1,999 · RTX 5090 (32 GB)48 GB+$4,650 · RTX A6000 (48 GB) · workstationSource: RunsOnMy Local LLM Hardware Index 2026-11-17, runsonmy.com/report/2026-11-17/ (CC BY 4.0)

SVG · 32 consumer cards with a known launch MSRP. Macs are not included because our table has no Mac prices.

This week vs last week

This is the first edition. Whole-catalog metrics as of the previous week, reconstructed by removing models released in the last 7 days. Download ranks for last week are not available, so top-60 metrics start week-over-week with the next edition.

MetricThis weekLast week (reconstructed)Change
Open models tracked298296+2
Median VRAM, all tracked (GB, 4-bit)8.89.5−0.7
All tracked models that run on 8 GB142141+1
All tracked models that run on 12 GB162160+2
All tracked models that run on 16 GB168166+2
All tracked models that run on 24 GB204202+2

New models this week

ModelReleasedParamsVRAM at 4-bit (est.)Smallest tier
SenseNova Loop 9B sensenova2026-11-179.4B6.4 GB8 GB
KORMo Web 11B KORMo-VL2026-11-1411B8.3 GB12 GB

VRAM by release quarter

Median VRAM of newly released models by quarterMedian VRAM of newly released models by quarterMedian GB at 4-bit, 8K context, of tracked models grouped by Hugging Face release date (n = models)24 GB41.32025 Q3n=28 · 46% fit 24 GB7.32025 Q4n=25 · 60% fit 24 GB8.62026 Q1n=24 · 79% fit 24 GB118.82026 Q2n=27 · 44% fit 24 GB73.52026 Q3n=33 · 45% fit 24 GB14.32026 Q4n=37 · 70% fit 24 GBTracked models skew to popular releases; small samples swing. Source: RunsOnMy Local LLM Hardware Index 2026-11-17 (CC BY 4.0)

SVG

The top 60 models

#ModelParams4-bit GB (est.)8-bit GB (est.)Smallest tier
1Qwen3 0.6B752M22.38 GB
2Qwen3 14B15B10.316.612 GB
3DeepSeek V3.2 MoE685B378.9665.6>48 GB
4Qwen3.8 27B28B17.628.524 GB
5Qwen3 VL 8B Instruct8.8B6.59.98 GB
6Qwen2.5 7B Instruct7.6B5.58.78 GB
7GLM 5.3 Flash MoE321B183.7318.4>48 GB
8Gemma 4 31B IT31B21.233.324 GB
9Qwen3.5 9B9.7B6.910.38 GB
10Qwen3 8B8.2B6.59.98 GB
11Unlimited OCR MoE3.3B2.948 GB
12gpt-oss 20B MoE21B11.812.316 GB
13Qwen3.5 2B2.3B2.32.98 GB
14Qwen2.5 VL 7B Instruct8.3B5.58.78 GB
15Gemma 4 26B A4B IT MoE26B17.326.424 GB
16Qwen2.5 1.5B Instruct1.5B1.92.78 GB
17Qwen3.5 4B4.7B3.95.48 GB
18Llama 3.1 8B Instruct8.0B6.29.68 GB
19Qwen2.5 0.5B Instruct494M1.21.48 GB
20Qwen3.6 35B A3B MoE36B21.836.224 GB
21Qwen3 32B33B21.135.124 GB
22Qwen3 VL 4B Instruct4.4B4.15.88 GB
23Qwen3 4B4.0B4.15.88 GB
24Qwen3.8 Flash Next MoE180B112.4176.4>48 GB
25Qwen3.5 0.8B873M1.51.78 GB
26Qwen3 1.7B2.0B2.83.68 GB
27Qwen3 4B Instruct 25074.0B4.15.88 GB
28Qwen3.6 27B28B18.128.524 GB
29gpt-oss 120B MoE117B59.660.1>48 GB
30DeepSeek V4 Flash 0731 MoE304B174.6302.4>48 GB
31Qwen3 Coder 30B A3B Instruct MoE31B18.831.724 GB
32Qwen2.5 3B Instruct3.1B2.94.38 GB
33Qwen2.5 VL 3B Instruct3.8B2.848 GB
34DeepSeek OCR MoE3.3B2.948 GB
35GLM 5.3 MoE753B430.4747.9>48 GB
36SmolVLM2 500M Video Instruct507M1.11.28 GB
37Qwen3.5 27B28B18.128.524 GB
38Qwen2.5 Coder 14B Instruct15B10.616.812 GB
39DeepSeek R1 MoE684B377.8665.5>48 GB
40DeepSeek V4.1 Flash MoE763B435.8756.6>48 GB
41Qwen2.5 Coder 7B Instruct7.6B5.58.78 GB
42Inkling MoE952B545800.8>48 GB
43Qwen3.5 35B A3B MoE36B21.836.224 GB
44DeepSeek V3 MoE684B377.8665.5>48 GB
45Llama 3.2 1B Instruct1.2B1.62.18 GB
46DeepSeek V4 Flash MoE291B167289.3>48 GB
47GLM 5 MoE754B426.1747.8>48 GB
48GLM 4.7 Flash MoE31B18.330.824 GB
49Mistral 7B Instruct v0.27.2B5.58.68 GB
50GLM 4.5 Air MoE110B70.6111.5>48 GB
51MiniMax M2.5 MoE229B131.6229.2>48 GB
52Gemma 3 4B IT4.3B3.558 GB
53Hy3 MoE299B172.8299>48 GB
54MiniCPM5 2B2.5B2.53.58 GB
55Llama 3.2 3B Instruct3.2B3.44.78 GB
56Kimi K3 MoE2.78T1584.12752.4>48 GB
57GLM 5.1 MoE754B429.6747.8>48 GB
58Llama 3.2 1B1.2B1.62.18 GB
59DiffusionGemma 26B A4B IT MoE26B17.126.424 GB
60HunyuanOCR1.1B22.58 GB

Method and caveats

Cite this

RunsOnMy (2026). Local LLM Hardware Index, edition 2026-11-17. https://runsonmy.com/report/2026-11-17/. Data CC BY 4.0.
BibTeX
@misc{runsonmy_index_20261117,
  title  = {Local LLM Hardware Index, edition 2026-11-17},
  author = {{RunsOnMy}},
  year   = {2026},
  url    = {https://runsonmy.com/report/2026-11-17/},
  note   = {Automated weekly snapshot; data CC BY 4.0}
}

Permalink: https://runsonmy.com/report/2026-11-17/ · Data: top60.csv, tiers.csv · Charts (SVG): vram-top60, cheapest-gpu, share-runnable, release-trend · License: CC BY 4.0 (credit "RunsOnMy" with a link).

All editions