Your rig

Radeon RX 6600 · 8 GB

128 GB system RAM · RDNA 2 · 224 GB/s · 7.0 GB usable VRAM
Edit hardware Export as JSON

Context length

KV cache at f16, 8K 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 Radeon RX 6600

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 Q5_K_M and 8K. The ordering is editorial; the sizes and speeds are computed. How the shortlists work.

RecommendationModelWhyQuant · totalTok/s est.
Best overall Qwen3.5 4B
Alibaba · Apache 2.0
The 6–8 GB pick: a real assistant in 3.4 GB of weights. Q5_K_M · 3.9 GB 44 Runs great
Best for coding Qwen3.5 4B
Alibaba · Apache 2.0
A coding agent in 3.4 GB — the 6–8 GB answer. Q5_K_M · 3.9 GB 44 Runs great
Best reasoning Qwen3.5 4B
Alibaba · Apache 2.0
Step-by-step reasoning in 8 GB. Q5_K_M · 3.9 GB 44 Runs great
Best for writing Qwen3.5 4B
Alibaba · Apache 2.0
Competent drafting in 8 GB. Q5_K_M · 3.9 GB 44 Runs great
Best vision Qwen3.5 4B
Alibaba · Apache 2.0
Reads screenshots in 8 GB. Q5_K_M · 3.9 GB 44 Runs great
Best for agents Qwen3.5 4B
Alibaba · Apache 2.0
Function calling in 8 GB. Q5_K_M · 3.9 GB 44 Runs great
Best translation Qwen3.5 4B
Alibaba · Apache 2.0
Usable translation in 8 GB. Q5_K_M · 3.9 GB 44 Runs great
Fastest good model Ling 3.0 Tiny 7.9B-A1.3B
inclusionAI · MIT
1.3B active with a latent cache on 6 of 24 layers. Q5_K_M · 5.9 GB 61 Runs great
Best long context Qwen3.5 2B
Alibaba · Apache 2.0
Long documents on 4 GB. Q5_K_M · 3.6 GB
at 128K context
90 Runs great

Everything that fits

Runs great

16 of 80 · fits with headroom for a longer prompt
ModelParamsQuantWeights +KV 8KTotal VRAM headroom Tok/sRuntime
Ling 3.0 Tiny 7.9B-A1.3B 7.9B MoE / 1.3B act Q5_K_M 5.2 GB 0.05 GB 5.9 GB
1.1 GB free
61 llama.cpp Run
Gemma 4 E2B 5.1B Q5_K_M 3.4 GB 0.07 GB 4.0 GB
3.0 GB free
40 Ollama Run
Qwen3.5 4B 4.66B Q5_K_M 3.1 GB 0.25 GB 3.9 GB
3.1 GB free
44 Ollama Run
Gemma 3 4B 4.3B Q5_K_M 2.8 GB 0.30 GB 3.7 GB
3.3 GB free
48 Ollama Run
Qwen3 4B 4.02B Q5_K_M 2.7 GB 1.13 GB 4.4 GB
2.6 GB free
51 Ollama Run
Ministral 3 3B 3.85B Q5_K_M 2.5 GB 0.81 GB 4.0 GB
3.0 GB free
53 Ollama Run
Phi-4-mini 3.8B 3.84B Q5_K_M 2.5 GB 1.00 GB 4.1 GB
2.9 GB free
53 Ollama Run
Granite 4.1 3B 3.4B Q5_K_M 2.2 GB 0.63 GB 3.5 GB
3.5 GB free
60 Ollama Run
Llama 3.2 3B Instruct 3.21B Q5_K_M 2.1 GB 0.88 GB 3.6 GB
3.4 GB free
64 Ollama Run
LFM2.5 2.6B 2.7B Q5_K_M 1.8 GB 0.13 GB 2.5 GB
4.5 GB free
76 llama.cpp Run
Qwen3.5 2B 2.27B Q5_K_M 1.5 GB 0.09 GB 2.2 GB
4.8 GB free
90 Ollama Run
Qwen3 1.7B 1.72B Q5_K_M 1.1 GB 0.88 GB 2.6 GB
4.4 GB free
119 Ollama Run
Llama 3.2 1B Instruct 1.24B Q5_K_M 0.8 GB 0.25 GB 1.7 GB
5.3 GB free
166 Ollama Run
Gemma 3 1B 1.0B Q5_K_M 0.7 GB 0.04 GB 1.3 GB
5.7 GB free
205 Ollama Run
Qwen3.5 0.8B 0.87B Q5_K_M 0.6 GB 0.09 GB 1.3 GB
5.7 GB free
236 Ollama Run
Qwen3 0.6B 0.6B Q5_K_M 0.4 GB 0.88 GB 1.9 GB
5.1 GB free
342 Ollama Run

Runs — tight

7 models · fits at 8K, drop context or quant to go longer
ModelParamsQuantWeights +KV 8KTotal VRAM headroom Tok/sRuntime
LFM2.5 8B-A1B 8.47B MoE / 1.5B act Q5_K_M 5.6 GB 0.09 GB 6.3 GB
0.7 GB free
53 llama.cpp Run
Fara 7B 8.29B Q5_K_M 5.5 GB 0.44 GB 6.5 GB
0.5 GB free
25 llama.cpp Run
Llama 3.1 8B Instruct 8.03B Q5_K_M 5.3 GB 1.00 GB 6.9 GB
0.1 GB free
26 llama.cpp Run
Gemma 4 E4B 8.0B Q5_K_M 5.3 GB 0.14 GB 6.0 GB
1.0 GB free
26 llama.cpp Run
DeepSeek-R1-Distill-Qwen 7B 7.62B Q5_K_M 5.0 GB 0.44 GB 6.1 GB
0.9 GB free
27 llama.cpp Run
Qwen2.5-Coder 7B 7.62B Q5_K_M 5.0 GB 0.44 GB 6.1 GB
0.9 GB free
27 llama.cpp Run
Mistral 7B Instruct v0.3 7.25B Q5_K_M 4.8 GB 1.00 GB 6.4 GB
0.6 GB free
28 llama.cpp Run

Won't fit in VRAM

57 models · some are possible with CPU offload
ModelParamsQuantWeights +KV 8KTotal Over budget Tok/sRuntime
Qwen3.5 122B-A10B 125B MoE / 10B act Q5_K_M 82.5 GB 0.19 GB 83.3 GB +76.3 GB ~2.2 llama.cpp Why
Nemotron 3 Super 120B-A12B 124B MoE / 12B act Q5_K_M 81.8 GB 0.06 GB 82.5 GB +75.5 GB ~1.9 llama.cpp Why
Ling 3.0 Flash 124B-A5B 124B MoE / 5.1B act Q5_K_M 81.8 GB 0.06 GB 82.5 GB +75.5 GB ~4.4 llama.cpp Why
Mistral Small 4 119B-A6B 119B MoE / 6.5B act Q5_K_M 78.5 GB 0.18 GB 79.3 GB +72.3 GB ~3.5 llama.cpp Why
gpt-oss 120B 117B MoE / 5.1B act MXFP4 60.5 GB 0.29 GB 61.4 GB +54.4 GB ~5.7 llama.cpp Why
Llama 4 Scout 109B-A17B 109B MoE / 17B act Q5_K_M 71.9 GB 1.50 GB 74.0 GB +67.0 GB ~1.3 llama.cpp Why
Llama 3.3 70B Instruct 70.6B Q5_K_M 46.6 GB 2.50 GB 49.7 GB +42.7 GB ~0.8 llama.cpp Why
DeepSeek-R1-Distill-Llama 70B 70.6B Q5_K_M 46.6 GB 2.50 GB 49.7 GB +42.7 GB ~0.8 llama.cpp Why
Qwen3.6 35B-A3B 35.9B MoE / 3.3B act Q5_K_M 23.7 GB 0.16 GB 24.5 GB +17.5 GB ~7.9 llama.cpp Why
Ornith 1.5 35B-A3B 35.9B MoE / 3.3B act Q5_K_M 23.7 GB 0.16 GB 24.5 GB +17.5 GB ~7.9 llama.cpp Why
LLM-jp 4 33B Thinking 33.2B Q5_K_M 21.9 GB 2.00 GB 24.5 GB +17.5 GB ~1.9 llama.cpp Why
Qwen3 32B 32.8B Q5_K_M 21.7 GB 2.00 GB 24.3 GB +17.3 GB ~2.0 llama.cpp Why
DeepSeek-R1-Distill-Qwen 32B 32.8B Q5_K_M 21.7 GB 2.00 GB 24.3 GB +17.3 GB ~2.0 llama.cpp Why
Qwen2.5-Coder 32B 32.8B Q5_K_M 21.7 GB 2.00 GB 24.3 GB +17.3 GB ~2.0 llama.cpp Why
Olmo 3.1 32B Instruct 32.2B Q5_K_M 21.3 GB 1.25 GB 23.1 GB +16.1 GB ~2.1 llama.cpp Why
Nemotron 3.5 Lightning 30B-A3B 31.6B MoE / 3.2B act Q5_K_M 20.9 GB 0.05 GB 21.5 GB +14.5 GB ~8.5 llama.cpp Why
Gemma 4 31B 31.3B Q5_K_M 20.7 GB 2.03 GB 23.3 GB +16.3 GB ~2.1 llama.cpp Why
GLM-4.7-Flash 30B-A3B 31.2B MoE / 3B act Q5_K_M 20.6 GB 0.41 GB 21.6 GB +14.6 GB ~9.0 llama.cpp Why
Qwen3 30B-A3B 30.5B MoE / 3.3B act Q5_K_M 20.1 GB 0.75 GB 21.5 GB +14.5 GB ~8.1 llama.cpp Why
Qwen3 Coder 30B-A3B 30.5B MoE / 3.3B act Q5_K_M 20.1 GB 0.75 GB 21.5 GB +14.5 GB ~8.1 llama.cpp Why
Muse Glimmer 30B 29.8B Q5_K_M 19.7 GB 0.18 GB 20.4 GB +13.4 GB ~2.4 llama.cpp Why
Granite 4.1 30B 28.9B Q5_K_M 19.1 GB 2.00 GB 21.7 GB +14.7 GB ~2.3 llama.cpp Why
Qwen3.8 27B 27.8B Q5_K_M 18.4 GB 0.50 GB 19.5 GB +12.5 GB ~2.6 llama.cpp Why
Qwen3.6 27B 27.8B Q5_K_M 18.4 GB 0.50 GB 19.5 GB +12.5 GB ~2.6 llama.cpp Why
Gemma 3 27B 27.4B Q5_K_M 18.1 GB 1.03 GB 19.7 GB +12.7 GB ~2.6 llama.cpp Why
Gemma 4 26B-A4B 26.5B MoE / 3.8B act Q5_K_M 17.5 GB 0.51 GB 18.6 GB +11.6 GB ~7.4 llama.cpp Why
Devstral Small 2 24B 24B Q5_K_M 15.8 GB 1.25 GB 17.7 GB +10.7 GB ~3.0 llama.cpp Why
Mistral Small 3.2 24B 23.6B Q5_K_M 15.6 GB 1.25 GB 17.4 GB +10.4 GB ~3.1 llama.cpp Why
gpt-oss 20B 20.9B MoE / 3.6B act MXFP4 10.8 GB 0.19 GB 11.6 GB +4.6 GB ~13 llama.cpp Why
Qwen3 14B 14.8B Q5_K_M 9.8 GB 1.25 GB 11.6 GB +4.6 GB ~6.1 llama.cpp Why
DeepSeek-R1-Distill-Qwen 14B 14.8B Q5_K_M 9.8 GB 1.50 GB 11.9 GB +4.9 GB ~5.9 llama.cpp Why
Qwen2.5-Coder 14B 14.8B Q5_K_M 9.8 GB 1.50 GB 11.9 GB +4.9 GB ~5.9 llama.cpp Why
Phi-4 14B 14.7B Q5_K_M 9.7 GB 1.56 GB 11.9 GB +4.9 GB ~5.9 llama.cpp Why
Ministral 3 14B 13.9B Q5_K_M 9.2 GB 1.25 GB 11.0 GB +4.0 GB ~6.7 llama.cpp Why
Gemma 3 12B 12.2B Q5_K_M 8.1 GB 0.81 GB 9.5 GB +2.5 GB ~9.2 llama.cpp Why
Mistral NeMo 12B 12.2B Q5_K_M 8.1 GB 1.25 GB 9.9 GB +2.9 GB ~8.5 llama.cpp Why
Gemma 4 12B 12B Q5_K_M 7.9 GB 0.81 GB 9.3 GB +2.3 GB ~9.5 llama.cpp Why
Qwen3.5 9B 9.65B Q5_K_M 6.4 GB 0.25 GB 7.2 GB +0.2 GB ~19 llama.cpp Why
Ornith 1.5 9B 9.41B Q5_K_M 6.2 GB 0.25 GB 7.1 GB +0.1 GB ~21 llama.cpp Why
Ministral 3 8B 8.92B Q5_K_M 5.9 GB 1.06 GB 7.6 GB +0.6 GB ~18 llama.cpp Why
Granite 4.1 8B 8.79B Q5_K_M 5.8 GB 1.25 GB 7.7 GB +0.7 GB ~18 llama.cpp Why
Qwen3 8B 8.19B Q5_K_M 5.4 GB 1.13 GB 7.1 GB +0.1 GB ~24 llama.cpp Why
Olmo 3 7B Instruct 7.3B Q5_K_M 4.8 GB 2.50 GB 7.9 GB +0.9 GB ~19 llama.cpp Why
Kimi K3 2.8T-A104B 2780B MoE / 104B act Q5_K_M 1835.0 GB 0.21 GB 1835.8 GB +1828.8 GB ~0.2 llama.cpp Why
Qwen3.8 2.4T-A95B 2446B MoE / 95B act Q5_K_M 1614.5 GB 0.72 GB 1615.9 GB +1608.9 GB ~0.2 llama.cpp Why
DeepSeek V4 Pro 1.6T-A49B 1650B MoE / 49B act Q5_K_M 1089.1 GB 0.54 GB 1090.3 GB +1083.3 GB ~0.4 llama.cpp Why
Kimi K2.6 1T-A32B 1027B MoE / 32B act Q5_K_M 677.9 GB 0.54 GB 679.0 GB +672.0 GB ~0.7 llama.cpp Why
GLM-5.2 744B-A40B 753B MoE / 40B act Q5_K_M 497.0 GB 0.69 GB 498.3 GB +491.3 GB ~0.5 llama.cpp Why
GLM-5.3 744B-A40B 753B MoE / 40B act Q5_K_M 497.0 GB 0.69 GB 498.3 GB +491.3 GB ~0.5 llama.cpp Why
DeepSeek-R1 671B 671B MoE / 37B act Q5_K_M 442.9 GB 0.54 GB 444.0 GB +437.0 GB ~0.6 llama.cpp Why
DeepSeek V4.1 Flash 552B-A16B New 552B MoE / 16B act Q5_K_M 364.4 GB 0.04 GB 365.0 GB +358.0 GB ~1.3 — Why
Qwen3.5 397B-A17B 403B MoE / 17B act Q5_K_M 266.0 GB 0.23 GB 266.8 GB +259.8 GB ~1.3 llama.cpp Why
Ornith 1.5 397B-A17B 397B MoE / 17B act Q5_K_M 262.0 GB 0.23 GB 262.9 GB +255.9 GB ~1.3 llama.cpp Why
GLM-5.3-Flash 320B-A18B 321B MoE / 18B act Q5_K_M 211.9 GB 0.09 GB 212.6 GB +205.6 GB ~1.2 llama.cpp Why
DeepSeek V4 Flash 284B-A13B 284B MoE / 13B act Q5_K_M 187.5 GB 0.38 GB 188.4 GB +181.4 GB ~1.7 llama.cpp Why
Qwen3 235B-A22B 235B MoE / 22B act Q5_K_M 155.1 GB 1.47 GB 157.2 GB +150.2 GB ~1.0 llama.cpp Why
Qwen3.8-Flash-Next 180B-A6B 180B MoE / 6B act Q5_K_M 118.8 GB 0.19 GB 119.6 GB +112.6 GB ~3.7 llama.cpp Why

Totals = quantised weights + f16 KV cache at 8K tokens + 0.6 GB runtime overhead, against 7.0 GB usable VRAM (8 GB less display and driver reserve). Tokens per second are estimated from 224 GB/s peak memory bandwidth at batch 1 — not measured. Read the method.


Same memory, different speed

10 of the 12 devices have the same 7.0 GB usable, so they run exactly the models above. Generation speed scales with memory bandwidth.

DeviceBandwidthSpeed vs yours
GeForce RTX 3070 Ti
Ampere · 2021
608 GB/s 2.71×
Arc A750
Alchemist · 2022
512 GB/s 2.29×
GeForce RTX 2080 Super
Turing · 2019
496 GB/s 2.21×
GeForce RTX 5060
Blackwell · 2025
448 GB/s 2.00×
GeForce RTX 3070
Ampere · 2020
448 GB/s 2.00×
GeForce RTX 3060 Ti
Ampere · 2020
448 GB/s 2.00×
GeForce RTX 4060 Ti
Ada Lovelace · 2023
288 GB/s 1.29×
GeForce RTX 4060
Ada Lovelace · 2023
272 GB/s 1.21×
GeForce RTX 4070 Laptop
Ada Lovelace · 2023
256 GB/s 1.14×
GeForce RTX 4060 Laptop
Ada Lovelace · 2023
256 GB/s 1.14×

Best LLM for 8 GB VRAM, with every card in the class →

What more memory unlocks

At Q4_K_M and 8K context this device fits 28 of 80 models. Each step up, and the largest models it adds:

Every device ranked by what it can run →