Your rig

Radeon RX 7600 XT · 16 GB

64 GB system RAM · RDNA 3 · 288 GB/s · 14.4 GB usable VRAM
Edit hardware Export as JSON

Context length

KV cache at f16, 4K 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 7600 XT

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 Q2_K and 4K. 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. Q2_K · 11.7 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. Q2_K · 11.7 GB 16 Runs great
Best reasoning Qwen3.8 27B
Alibaba · Apache 2.0
Thinking mode plus the best agentic reasoning at 24 GB. Q2_K · 11.7 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. Q2_K · 11.3 GB 45 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. Q2_K · 11.7 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). Q2_K · 11.7 GB 16 Runs great
Best translation Qwen3.8 27B
Alibaba · Apache 2.0
119 languages, and Qwen translations read as idiomatic rather than literal. Q2_K · 11.7 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. Q2_K · 11.3 GB 45 Runs great
Best long context Nemotron 3.5 Lightning 30B-A3B
NVIDIA · OpenMDW-1.1
262K, and a Mamba hybrid whose cache barely grows. Q2_K · 13.7 GB
at 128K context
54 Tight fit

Everything that fits

Runs great

44 of 80 · fits with headroom for a longer prompt
ModelParamsQuantWeights +KV 4KTotal VRAM headroom Tok/sRuntime
Qwen3.8 27B 27.8B Q2_K 10.8 GB 0.25 GB 11.7 GB
2.7 GB free
16 Ollama Run
Qwen3.6 27B 27.8B Q2_K 10.8 GB 0.25 GB 11.7 GB
2.7 GB free
16 Ollama Run
Gemma 3 27B 27.4B Q2_K 10.7 GB 0.72 GB 12.0 GB
2.4 GB free
16 Ollama Run
Gemma 4 26B-A4B 26.5B MoE / 3.8B act Q2_K 10.3 GB 0.35 GB 11.3 GB
3.1 GB free
45 Ollama Run
Devstral Small 2 24B 24B Q2_K 9.4 GB 0.63 GB 10.6 GB
3.8 GB free
19 llama.cpp Run
Mistral Small 3.2 24B 23.6B Q2_K 9.2 GB 0.63 GB 10.4 GB
4.0 GB free
19 Ollama Run
gpt-oss 20B 20.9B MoE / 3.6B act MXFP4 10.8 GB 0.10 GB 11.5 GB
2.9 GB free
36 Ollama Run
Qwen3 14B 14.8B Q2_K 5.8 GB 0.63 GB 7.0 GB
7.4 GB free
30 Ollama Run
DeepSeek-R1-Distill-Qwen 14B 14.8B Q2_K 5.8 GB 0.75 GB 7.1 GB
7.3 GB free
30 Ollama Run
Qwen2.5-Coder 14B 14.8B Q2_K 5.8 GB 0.75 GB 7.1 GB
7.3 GB free
30 Ollama Run
Phi-4 14B 14.7B Q2_K 5.7 GB 0.78 GB 7.1 GB
7.3 GB free
30 Ollama Run
Ministral 3 14B 13.9B Q2_K 5.4 GB 0.63 GB 6.6 GB
7.8 GB free
32 Ollama Run
Gemma 3 12B 12.2B Q2_K 4.8 GB 0.56 GB 5.9 GB
8.5 GB free
37 Ollama Run
Mistral NeMo 12B 12.2B Q2_K 4.8 GB 0.63 GB 6.0 GB
8.4 GB free
37 Ollama Run
Gemma 4 12B 12B Q2_K 4.7 GB 0.56 GB 5.8 GB
8.6 GB free
37 Ollama Run
Qwen3.5 9B 9.65B Q2_K 3.8 GB 0.13 GB 4.5 GB
9.9 GB free
46 Ollama Run
Ornith 1.5 9B 9.41B Q2_K 3.7 GB 0.13 GB 4.4 GB
10.0 GB free
48 Ollama Run
Ministral 3 8B 8.92B Q2_K 3.5 GB 0.53 GB 4.6 GB
9.8 GB free
50 Ollama Run
Granite 4.1 8B 8.79B Q2_K 3.4 GB 0.63 GB 4.7 GB
9.7 GB free
51 Ollama Run
LFM2.5 8B-A1B 8.47B MoE / 1.5B act Q2_K 3.3 GB 0.05 GB 4.0 GB
10.4 GB free
115 Ollama Run
Fara 7B 8.29B Q2_K 3.2 GB 0.22 GB 4.1 GB
10.3 GB free
54 llama.cpp Run
Qwen3 8B 8.19B Q2_K 3.2 GB 0.56 GB 4.4 GB
10.0 GB free
55 Ollama Run
Llama 3.1 8B Instruct 8.03B Q2_K 3.1 GB 0.50 GB 4.2 GB
10.2 GB free
56 Ollama Run
Gemma 4 E4B 8.0B Q2_K 3.1 GB 0.09 GB 3.8 GB
10.6 GB free
56 Ollama Run
Ling 3.0 Tiny 7.9B-A1.3B 7.9B MoE / 1.3B act Q2_K 3.1 GB 0.03 GB 3.7 GB
10.7 GB free
132 llama.cpp Run
DeepSeek-R1-Distill-Qwen 7B 7.62B Q2_K 3.0 GB 0.22 GB 3.8 GB
10.6 GB free
59 Ollama Run
Qwen2.5-Coder 7B 7.62B Q2_K 3.0 GB 0.22 GB 3.8 GB
10.6 GB free
59 Ollama Run
Olmo 3 7B Instruct 7.3B Q2_K 2.8 GB 2.00 GB 5.4 GB
9.0 GB free
61 Ollama Run
Mistral 7B Instruct v0.3 7.25B Q2_K 2.8 GB 0.50 GB 3.9 GB
10.5 GB free
62 Ollama Run
Gemma 4 E2B 5.1B Q2_K 2.0 GB 0.04 GB 2.6 GB
11.8 GB free
88 Ollama Run
Qwen3.5 4B 4.66B Q2_K 1.8 GB 0.13 GB 2.5 GB
11.9 GB free
96 Ollama Run
Gemma 3 4B 4.3B Q2_K 1.7 GB 0.20 GB 2.5 GB
11.9 GB free
104 Ollama Run
Qwen3 4B 4.02B Q2_K 1.6 GB 0.56 GB 2.7 GB
11.7 GB free
111 Ollama Run
Ministral 3 3B 3.85B Q2_K 1.5 GB 0.41 GB 2.5 GB
11.9 GB free
116 Ollama Run
Phi-4-mini 3.8B 3.84B Q2_K 1.5 GB 0.50 GB 2.6 GB
11.8 GB free
116 Ollama Run
Granite 4.1 3B 3.4B Q2_K 1.3 GB 0.31 GB 2.2 GB
12.2 GB free
131 Ollama Run
Llama 3.2 3B Instruct 3.21B Q2_K 1.3 GB 0.44 GB 2.3 GB
12.1 GB free
139 Ollama Run
LFM2.5 2.6B 2.7B Q2_K 1.1 GB 0.06 GB 1.7 GB
12.7 GB free
166 llama.cpp Run
Qwen3.5 2B 2.27B Q2_K 0.9 GB 0.05 GB 1.5 GB
12.9 GB free
197 Ollama Run
Qwen3 1.7B 1.72B Q2_K 0.7 GB 0.44 GB 1.7 GB
12.7 GB free
260 Ollama Run
Llama 3.2 1B Instruct 1.24B Q2_K 0.5 GB 0.13 GB 1.2 GB
13.2 GB free
361 Ollama Run
Gemma 3 1B 1.0B Q2_K 0.4 GB 0.03 GB 1.0 GB
13.4 GB free
447 Ollama Run
Qwen3.5 0.8B 0.87B Q2_K 0.3 GB 0.05 GB 1.0 GB
13.4 GB free
514 Ollama Run
Qwen3 0.6B 0.6B Q2_K 0.2 GB 0.44 GB 1.3 GB
13.1 GB free
745 Ollama Run

Runs — tight

11 models · fits at 4K, drop context or quant to go longer
ModelParamsQuantWeights +KV 4KTotal VRAM headroom Tok/sRuntime
Qwen3 32B 32.8B Q2_K 12.8 GB 1.00 GB 14.4 GB
0.0 GB free
14 llama.cpp Run
DeepSeek-R1-Distill-Qwen 32B 32.8B Q2_K 12.8 GB 1.00 GB 14.4 GB
0.0 GB free
14 llama.cpp Run
Qwen2.5-Coder 32B 32.8B Q2_K 12.8 GB 1.00 GB 14.4 GB
0.0 GB free
14 llama.cpp Run
Olmo 3.1 32B Instruct 32.2B Q2_K 12.6 GB 1.00 GB 14.2 GB
0.2 GB free
14 llama.cpp Run
Nemotron 3.5 Lightning 30B-A3B 31.6B MoE / 3.2B act Q2_K 12.3 GB 0.02 GB 12.9 GB
1.5 GB free
54 llama.cpp Run
Gemma 4 31B 31.3B Q2_K 12.2 GB 1.41 GB 14.2 GB
0.2 GB free
14 llama.cpp Run
GLM-4.7-Flash 30B-A3B 31.2B MoE / 3B act Q2_K 12.2 GB 0.21 GB 13.0 GB
1.4 GB free
57 llama.cpp Run
Qwen3 30B-A3B 30.5B MoE / 3.3B act Q2_K 11.9 GB 0.38 GB 12.9 GB
1.5 GB free
52 llama.cpp Run
Qwen3 Coder 30B-A3B 30.5B MoE / 3.3B act Q2_K 11.9 GB 0.38 GB 12.9 GB
1.5 GB free
52 llama.cpp Run
Muse Glimmer 30B 29.8B Q2_K 11.6 GB 0.13 GB 12.3 GB
2.1 GB free
15 llama.cpp Run
Granite 4.1 30B 28.9B Q2_K 11.3 GB 1.00 GB 12.9 GB
1.5 GB free
15 llama.cpp Run

Won't fit in VRAM

25 models · some are possible with CPU offload
ModelParamsQuantWeights +KV 4KTotal Over budget Tok/sRuntime
Qwen3.5 122B-A10B 125B MoE / 10B act Q2_K 48.7 GB 0.09 GB 49.4 GB +35.0 GB ~4.6 llama.cpp Why
Nemotron 3 Super 120B-A12B 124B MoE / 12B act Q2_K 48.4 GB 0.03 GB 49.0 GB +34.6 GB ~3.9 llama.cpp Why
Ling 3.0 Flash 124B-A5B 124B MoE / 5.1B act Q2_K 48.4 GB 0.03 GB 49.0 GB +34.6 GB ~9.1 llama.cpp Why
Mistral Small 4 119B-A6B 119B MoE / 6.5B act Q2_K 46.4 GB 0.09 GB 47.1 GB +32.7 GB ~7.2 llama.cpp Why
gpt-oss 120B 117B MoE / 5.1B act MXFP4 60.5 GB 0.15 GB 61.3 GB +46.9 GB ~6.4 llama.cpp Why
Llama 4 Scout 109B-A17B 109B MoE / 17B act Q2_K 42.5 GB 0.75 GB 43.9 GB +29.5 GB ~2.8 llama.cpp Why
Llama 3.3 70B Instruct 70.6B Q2_K 27.5 GB 1.25 GB 29.4 GB +15.0 GB ~2.1 llama.cpp Why
DeepSeek-R1-Distill-Llama 70B 70.6B Q2_K 27.5 GB 1.25 GB 29.4 GB +15.0 GB ~2.1 llama.cpp Why
Qwen3.6 35B-A3B 35.9B MoE / 3.3B act Q2_K 14.0 GB 0.08 GB 14.7 GB +0.3 GB ~48 llama.cpp Why
Ornith 1.5 35B-A3B 35.9B MoE / 3.3B act Q2_K 14.0 GB 0.08 GB 14.7 GB +0.3 GB ~48 llama.cpp Why
LLM-jp 4 33B Thinking 33.2B Q2_K 12.9 GB 1.00 GB 14.5 GB +0.1 GB ~13 llama.cpp Why
Kimi K3 2.8T-A104B 2780B MoE / 104B act Q2_K 1084.2 GB 0.11 GB 1084.9 GB +1070.5 GB ~0.3 llama.cpp Why
Qwen3.8 2.4T-A95B 2446B MoE / 95B act Q2_K 953.9 GB 0.36 GB 954.9 GB +940.5 GB ~0.4 llama.cpp Why
DeepSeek V4 Pro 1.6T-A49B 1650B MoE / 49B act Q2_K 643.5 GB 0.27 GB 644.4 GB +630.0 GB ~0.7 llama.cpp Why
Kimi K2.6 1T-A32B 1027B MoE / 32B act Q2_K 400.5 GB 0.27 GB 401.4 GB +387.0 GB ~1.2 llama.cpp Why
GLM-5.2 744B-A40B 753B MoE / 40B act Q2_K 293.7 GB 0.34 GB 294.6 GB +280.2 GB ~0.9 llama.cpp Why
GLM-5.3 744B-A40B 753B MoE / 40B act Q2_K 293.7 GB 0.34 GB 294.6 GB +280.2 GB ~0.9 llama.cpp Why
DeepSeek-R1 671B 671B MoE / 37B act Q2_K 261.7 GB 0.27 GB 262.6 GB +248.2 GB ~1.0 llama.cpp Why
DeepSeek V4.1 Flash 552B-A16B New 552B MoE / 16B act Q2_K 215.3 GB 0.02 GB 215.9 GB +201.5 GB ~2.4 — Why
Qwen3.5 397B-A17B 403B MoE / 17B act Q2_K 157.2 GB 0.12 GB 157.9 GB +143.5 GB ~2.3 llama.cpp Why
Ornith 1.5 397B-A17B 397B MoE / 17B act Q2_K 154.8 GB 0.12 GB 155.5 GB +141.1 GB ~2.3 llama.cpp Why
GLM-5.3-Flash 320B-A18B 321B MoE / 18B act Q2_K 125.2 GB 0.04 GB 125.8 GB +111.4 GB ~2.2 llama.cpp Why
DeepSeek V4 Flash 284B-A13B 284B MoE / 13B act Q2_K 110.8 GB 0.19 GB 111.5 GB +97.1 GB ~3.1 llama.cpp Why
Qwen3 235B-A22B 235B MoE / 22B act Q2_K 91.6 GB 0.73 GB 93.0 GB +78.6 GB ~1.8 llama.cpp Why
Qwen3.8-Flash-Next 180B-A6B 180B MoE / 6B act Q2_K 70.2 GB 0.09 GB 70.9 GB +56.5 GB ~7.1 llama.cpp Why

Totals = quantised weights + f16 KV cache at 4K tokens + 0.6 GB runtime overhead, against 14.4 GB usable VRAM (16 GB less display and driver reserve). Tokens per second are estimated from 288 GB/s peak memory bandwidth at batch 1 — not measured. Read the method.


Same memory, different speed

10 of the 19 devices have the same 14.4 GB usable, so they run exactly the models above. Generation speed scales with memory bandwidth.

DeviceBandwidthSpeed vs yours
GeForce RTX 5080
Blackwell · 2025
960 GB/s 3.33×
GeForce RTX 5070 Ti
Blackwell · 2025
896 GB/s 3.11×
CMP 100-210
Volta · 2021
829 GB/s 2.88×
GeForce RTX 4080 Super
Ada Lovelace · 2024
736 GB/s 2.56×
Tesla P100 16 GB
Pascal · 2016
732 GB/s 2.54×
GeForce RTX 4080
Ada Lovelace · 2022
717 GB/s 2.49×
GeForce RTX 4070 Ti Super
Ada Lovelace · 2024
672 GB/s 2.33×
Radeon RX 9070 XT
RDNA 4 · 2025
645 GB/s 2.24×
Radeon RX 9070
RDNA 4 · 2025
645 GB/s 2.24×
Radeon RX 7800 XT
RDNA 3 · 2023
624 GB/s 2.17×

Best LLM for 16 GB VRAM, with every card in the class →

What more memory unlocks

At Q4_K_M and 8K context this device fits 38 of 80 models. Each step up, and the largest models it adds:

Every device ranked by what it can run →