Your rig

RTX 5000 Ada · 32 GB

128 GB system RAM · Ada Lovelace · 576 GB/s · 30.4 GB usable VRAM
Edit hardware Export as JSON

Context length

KV cache at f16, 32K tokens

Quality floor
Q8_0 — near-lossless Q6_K Q5_K_M Q4_K_M — recommended Q3_K_M Q2_K — damaged
KV cache type
System RAM (GB)
Runtime
Ollamallama.cppLM StudiovLLM

4 of 5 runtimes drive this device.

Allow CPU offload Include partial-offload fits

Offloaded layers run at 60 GB/s host bandwidth — expect single-digit tokens per second.

Which one should I actually run?

Best models for your RTX 5000 Ada

Updated 8 Oct 2026 1 new this month

What do you want to do? Each row is the best-ranked model on that use case's shortlist that fits at Q6_K and 32K. The ordering is editorial; the sizes and speeds are computed. How the shortlists work.

RecommendationModelWhyQuant · totalTok/s est.
Best overall Qwen3.8 27B
Alibaba · Apache 2.0
Best capability per gigabyte: a current-generation dense 27B with vision and 262K context. Q6_K · 23.8 GB 16 Runs great
Best for coding Qwen3.8 27B
Alibaba · Apache 2.0
Terminal-Bench 73, DeepSWE 42 — a generation ahead of anything else that fits 24 GB. Q6_K · 23.8 GB 16 Runs great
Best reasoning Qwen3.8 27B
Alibaba · Apache 2.0
Thinking mode plus the best agentic reasoning at 24 GB. Q6_K · 23.8 GB 16 Runs great
Best for writing Gemma 4 26B-A4B
Google · Apache 2.0
Gemma 4 prose at MoE speed, with room for a long draft in context. Q6_K · 22.3 GB 46 Runs great
Best vision Qwen3.8 27B
Alibaba · Apache 2.0
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. Q6_K · 23.8 GB 16 Runs great
Best for agents Qwen3.8 27B
Alibaba · Apache 2.0
Computer use and tool calling are what 3.8 was trained for (OSWorld 84). Q6_K · 23.8 GB 16 Runs great
Best translation Qwen3.8 27B
Alibaba · Apache 2.0
119 languages, and Qwen translations read as idiomatic rather than literal. Q6_K · 23.8 GB 16 Runs great
Fastest good model Gemma 4 26B-A4B
Google · Apache 2.0
3.8B active and a small file — the quickest Gemma 4 with vision. Q6_K · 22.3 GB 46 Runs great
Best long context Qwen3.8 27B
Alibaba · Apache 2.0
262K native, and only 16 of 64 blocks keep a KV cache — long context is cheap here. Q6_K · 29.8 GB
at 128K context
16 Tight fit

Everything that fits

Runs great

46 of 80 · fits with headroom for a longer prompt
ModelParamsQuantWeights +KV 32KTotal VRAM headroom Tok/sRuntime
Nemotron 3.5 Lightning 30B-A3B 31.6B MoE / 3.2B act Q6_K 24.1 GB 0.19 GB 24.9 GB
5.5 GB free
55 Ollama Run
Muse Glimmer 30B 29.8B Q6_K 22.8 GB 0.48 GB 23.8 GB
6.6 GB free
15 Ollama Run
Qwen3.8 27B 27.8B Q6_K 21.2 GB 2.00 GB 23.8 GB
6.6 GB free
16 Ollama Run
Qwen3.6 27B 27.8B Q6_K 21.2 GB 2.00 GB 23.8 GB
6.6 GB free
16 Ollama Run
Gemma 3 27B 27.4B Q6_K 20.9 GB 2.91 GB 24.4 GB
6.0 GB free
17 Ollama Run
Gemma 4 26B-A4B 26.5B MoE / 3.8B act Q6_K 20.2 GB 1.45 GB 22.3 GB
8.1 GB free
46 Ollama Run
Devstral Small 2 24B 24B Q6_K 18.3 GB 5.00 GB 23.9 GB
6.5 GB free
19 llama.cpp Run
Mistral Small 3.2 24B 23.6B Q6_K 18.0 GB 5.00 GB 23.6 GB
6.8 GB free
19 Ollama Run
gpt-oss 20B 20.9B MoE / 3.6B act MXFP4 10.8 GB 0.75 GB 12.2 GB
18.2 GB free
72 Ollama Run
Qwen3 14B 14.8B Q6_K 11.3 GB 5.00 GB 16.9 GB
13.5 GB free
31 Ollama Run
DeepSeek-R1-Distill-Qwen 14B 14.8B Q6_K 11.3 GB 6.00 GB 17.9 GB
12.5 GB free
31 Ollama Run
Qwen2.5-Coder 14B 14.8B Q6_K 11.3 GB 6.00 GB 17.9 GB
12.5 GB free
31 Ollama Run
Phi-4 14B 14.7B Q6_K 11.2 GB 3.13 GB 15.0 GB
15.4 GB free
31 Ollama Run
Ministral 3 14B 13.9B Q6_K 10.6 GB 5.00 GB 16.2 GB
14.2 GB free
33 Ollama Run
Gemma 3 12B 12.2B Q6_K 9.3 GB 2.31 GB 12.2 GB
18.2 GB free
37 Ollama Run
Mistral NeMo 12B 12.2B Q6_K 9.3 GB 5.00 GB 14.9 GB
15.5 GB free
37 Ollama Run
Gemma 4 12B 12B Q6_K 9.2 GB 2.31 GB 12.1 GB
18.3 GB free
38 Ollama Run
Qwen3.5 9B 9.65B Q6_K 7.4 GB 1.00 GB 9.0 GB
21.4 GB free
47 Ollama Run
Ornith 1.5 9B 9.41B Q6_K 7.2 GB 1.00 GB 8.8 GB
21.6 GB free
49 Ollama Run
Ministral 3 8B 8.92B Q6_K 6.8 GB 4.25 GB 11.7 GB
18.7 GB free
51 Ollama Run
Granite 4.1 8B 8.79B Q6_K 6.7 GB 5.00 GB 12.3 GB
18.1 GB free
52 Ollama Run
LFM2.5 8B-A1B 8.47B MoE / 1.5B act Q6_K 6.5 GB 0.38 GB 7.4 GB
23.0 GB free
117 Ollama Run
Fara 7B 8.29B Q6_K 6.3 GB 1.75 GB 8.7 GB
21.7 GB free
55 llama.cpp Run
Qwen3 8B 8.19B Q6_K 6.3 GB 4.50 GB 11.4 GB
19.0 GB free
56 Ollama Run
Llama 3.1 8B Instruct 8.03B Q6_K 6.1 GB 4.00 GB 10.7 GB
19.7 GB free
57 Ollama Run
Gemma 4 E4B 8.0B Q6_K 6.1 GB 0.47 GB 7.2 GB
23.2 GB free
57 Ollama Run
Ling 3.0 Tiny 7.9B-A1.3B 7.9B MoE / 1.3B act Q6_K 6.0 GB 0.21 GB 6.8 GB
23.6 GB free
135 llama.cpp Run
DeepSeek-R1-Distill-Qwen 7B 7.62B Q6_K 5.8 GB 1.75 GB 8.2 GB
22.2 GB free
60 Ollama Run
Qwen2.5-Coder 7B 7.62B Q6_K 5.8 GB 1.75 GB 8.2 GB
22.2 GB free
60 Ollama Run
Olmo 3 7B Instruct 7.3B Q6_K 5.6 GB 5.50 GB 11.7 GB
18.7 GB free
63 Ollama Run
Mistral 7B Instruct v0.3 7.25B Q6_K 5.5 GB 4.00 GB 10.1 GB
20.3 GB free
63 Ollama Run
Gemma 4 E2B 5.1B Q6_K 3.9 GB 0.23 GB 4.7 GB
25.7 GB free
90 Ollama Run
Qwen3.5 4B 4.66B Q6_K 3.6 GB 1.00 GB 5.2 GB
25.2 GB free
98 Ollama Run
Gemma 3 4B 4.3B Q6_K 3.3 GB 0.86 GB 4.7 GB
25.7 GB free
106 Ollama Run
Qwen3 4B 4.02B Q6_K 3.1 GB 4.50 GB 8.2 GB
22.2 GB free
114 Ollama Run
Ministral 3 3B 3.85B Q6_K 2.9 GB 3.25 GB 6.8 GB
23.6 GB free
119 Ollama Run
Phi-4-mini 3.8B 3.84B Q6_K 2.9 GB 4.00 GB 7.5 GB
22.9 GB free
119 Ollama Run
Granite 4.1 3B 3.4B Q6_K 2.6 GB 2.50 GB 5.7 GB
24.7 GB free
134 Ollama Run
Llama 3.2 3B Instruct 3.21B Q6_K 2.5 GB 3.50 GB 6.6 GB
23.8 GB free
142 Ollama Run
LFM2.5 2.6B 2.7B Q6_K 2.1 GB 0.50 GB 3.2 GB
27.2 GB free
169 llama.cpp Run
Qwen3.5 2B 2.27B Q6_K 1.7 GB 0.38 GB 2.7 GB
27.7 GB free
201 Ollama Run
Qwen3 1.7B 1.72B Q6_K 1.3 GB 3.50 GB 5.4 GB
25.0 GB free
265 Ollama Run
Llama 3.2 1B Instruct 1.24B Q6_K 0.9 GB 1.00 GB 2.5 GB
27.9 GB free
368 Ollama Run
Gemma 3 1B 1.0B Q6_K 0.8 GB 0.14 GB 1.5 GB
28.9 GB free
457 Ollama Run
Qwen3.5 0.8B 0.87B Q6_K 0.7 GB 0.38 GB 1.6 GB
28.8 GB free
525 Ollama Run
Qwen3 0.6B 0.6B Q6_K 0.5 GB 3.50 GB 4.6 GB
25.8 GB free
761 Ollama Run

Runs — tight

7 models · fits at 32K, drop context or quant to go longer
ModelParamsQuantWeights +KV 32KTotal VRAM headroom Tok/sRuntime
Qwen3.6 35B-A3B 35.9B MoE / 3.3B act Q6_K 27.4 GB 0.63 GB 28.6 GB
1.8 GB free
53 llama.cpp Run
Ornith 1.5 35B-A3B 35.9B MoE / 3.3B act Q6_K 27.4 GB 0.63 GB 28.6 GB
1.8 GB free
53 llama.cpp Run
Olmo 3.1 32B Instruct 32.2B Q6_K 24.6 GB 2.75 GB 27.9 GB
2.5 GB free
14 llama.cpp Run
Gemma 4 31B 31.3B Q6_K 23.9 GB 5.78 GB 30.3 GB
0.1 GB free
15 llama.cpp Run
GLM-4.7-Flash 30B-A3B 31.2B MoE / 3B act Q6_K 23.8 GB 1.65 GB 26.1 GB
4.3 GB free
59 llama.cpp Run
Qwen3 30B-A3B 30.5B MoE / 3.3B act Q6_K 23.3 GB 3.00 GB 26.9 GB
3.5 GB free
53 llama.cpp Run
Qwen3 Coder 30B-A3B 30.5B MoE / 3.3B act Q6_K 23.3 GB 3.00 GB 26.9 GB
3.5 GB free
53 llama.cpp Run

Won't fit in VRAM

27 models · some are possible with CPU offload
ModelParamsQuantWeights +KV 32KTotal Over budget Tok/sRuntime
Qwen3.8-Flash-Next 180B-A6B 180B MoE / 6B act Q6_K 137.5 GB 0.75 GB 138.8 GB +108.4 GB ~3.8 llama.cpp Why
Qwen3.5 122B-A10B 125B MoE / 10B act Q6_K 95.5 GB 0.75 GB 96.8 GB +66.4 GB ~2.5 llama.cpp Why
Nemotron 3 Super 120B-A12B 124B MoE / 12B act Q6_K 94.7 GB 0.25 GB 95.5 GB +65.1 GB ~2.1 llama.cpp Why
Ling 3.0 Flash 124B-A5B 124B MoE / 5.1B act Q6_K 94.7 GB 0.25 GB 95.5 GB +65.1 GB ~5.0 llama.cpp Why
Mistral Small 4 119B-A6B 119B MoE / 6.5B act Q6_K 90.9 GB 0.70 GB 92.2 GB +61.8 GB ~3.9 llama.cpp Why
gpt-oss 120B 117B MoE / 5.1B act MXFP4 60.5 GB 1.13 GB 62.2 GB +31.8 GB ~9.2 llama.cpp Why
Llama 4 Scout 109B-A17B 109B MoE / 17B act Q6_K 83.2 GB 2.63 GB 86.5 GB +56.1 GB ~1.5 llama.cpp Why
Llama 3.3 70B Instruct 70.6B Q6_K 53.9 GB 10.00 GB 64.5 GB +34.1 GB ~1.0 llama.cpp Why
DeepSeek-R1-Distill-Llama 70B 70.6B Q6_K 53.9 GB 10.00 GB 64.5 GB +34.1 GB ~1.0 llama.cpp Why
LLM-jp 4 33B Thinking 33.2B Q6_K 25.4 GB 8.00 GB 34.0 GB +3.6 GB ~6.2 llama.cpp Why
Qwen3 32B 32.8B Q6_K 25.0 GB 8.00 GB 33.6 GB +3.2 GB ~6.6 llama.cpp Why
DeepSeek-R1-Distill-Qwen 32B 32.8B Q6_K 25.0 GB 8.00 GB 33.6 GB +3.2 GB ~6.6 llama.cpp Why
Qwen2.5-Coder 32B 32.8B Q6_K 25.0 GB 8.00 GB 33.6 GB +3.2 GB ~6.6 llama.cpp Why
Granite 4.1 30B 28.9B Q6_K 22.1 GB 8.00 GB 30.7 GB +0.3 GB ~14 llama.cpp Why
Kimi K3 2.8T-A104B 2780B MoE / 104B act Q6_K 2123.0 GB 0.84 GB 2124.5 GB +2094.1 GB ~0.2 llama.cpp Why
Qwen3.8 2.4T-A95B 2446B MoE / 95B act Q6_K 1868.0 GB 2.88 GB 1871.4 GB +1841.0 GB ~0.2 llama.cpp Why
DeepSeek V4 Pro 1.6T-A49B 1650B MoE / 49B act Q6_K 1260.1 GB 2.14 GB 1262.8 GB +1232.4 GB ~0.4 llama.cpp Why
Kimi K2.6 1T-A32B 1027B MoE / 32B act Q6_K 784.3 GB 2.14 GB 787.0 GB +756.6 GB ~0.6 llama.cpp Why
GLM-5.2 744B-A40B 753B MoE / 40B act Q6_K 575.1 GB 2.74 GB 578.4 GB +548.0 GB ~0.5 llama.cpp Why
GLM-5.3 744B-A40B 753B MoE / 40B act Q6_K 575.1 GB 2.74 GB 578.4 GB +548.0 GB ~0.5 llama.cpp Why
DeepSeek-R1 671B 671B MoE / 37B act Q6_K 512.4 GB 2.14 GB 515.2 GB +484.8 GB ~0.5 llama.cpp Why
DeepSeek V4.1 Flash 552B-A16B New 552B MoE / 16B act Q6_K 421.6 GB 0.15 GB 422.3 GB +391.9 GB ~1.2 vLLM Why
Qwen3.5 397B-A17B 403B MoE / 17B act Q6_K 307.8 GB 0.94 GB 309.3 GB +278.9 GB ~1.2 llama.cpp Why
Ornith 1.5 397B-A17B 397B MoE / 17B act Q6_K 303.2 GB 0.94 GB 304.7 GB +274.3 GB ~1.2 llama.cpp Why
GLM-5.3-Flash 320B-A18B 321B MoE / 18B act Q6_K 245.1 GB 0.34 GB 246.1 GB +215.7 GB ~1.1 llama.cpp Why
DeepSeek V4 Flash 284B-A13B 284B MoE / 13B act Q6_K 216.9 GB 1.51 GB 219.0 GB +188.6 GB ~1.6 llama.cpp Why
Qwen3 235B-A22B 235B MoE / 22B act Q6_K 179.5 GB 5.88 GB 185.9 GB +155.5 GB ~0.9 llama.cpp Why

Totals = quantised weights + f16 KV cache at 32K tokens + 0.6 GB runtime overhead, against 30.4 GB usable VRAM (32 GB less display and driver reserve). Tokens per second are estimated from 576 GB/s peak memory bandwidth at batch 1 — not measured. Read the method.


Same memory, different speed

These devices have the same 30.4 GB usable, so they run exactly the models above. Generation speed scales with memory bandwidth.

DeviceBandwidthSpeed vs yours
GeForce RTX 5090
Blackwell · 2025
1792 GB/s 3.11×
Radeon Pro W7800
RDNA 3 · 2023
576 GB/s same

Best LLM for 32 GB VRAM, with every card in the class →

What more memory unlocks

At Q4_K_M and 8K context this device fits 58 of 80 models. Each step up, and the largest models it adds:

Step upModels fitAdds, largest first
48 GB card 60 (+2) Llama 3.3 70B Instruct, DeepSeek-R1-Distill-Llama 70B

Every device ranked by what it can run →