What AI models run on the RX 7600?
The RX 7600 has 8 GB of VRAM (288 GB/s). About 7.6 GB is usable for a model. 161 of 298 open models we track fit at 8K context.
This card has 8 GB of VRAM. See every model that fits in 8 GB VRAM. Other 8 GB hardware: RTX 3060 Ti, RTX 3070, RTX 4060, RTX 4060 Ti 8GB, RTX 5060.
Compare for local LLMs: RX 7600 vs RTX 4060 all comparisons
Using a specific tool? Ollama models that run on the RX 7600 · LM Studio · llama.cpp offload
Models that fit, largest first
Bigger models with CPU offload
These popular models don't fit in the RX 7600's VRAM at 4-bit, but run with llama.cpp offload, keeping part of the model in system RAM. Speed range is DDR4-3200 to DDR5-6000 (estimate). Method.
| Model | Params | Offload plan |
|---|---|---|
| Qwen3.8 27B | 28B | Offload Q4_K_M, 32 GB RAM · ~1.9–3.8 tok/s |
| GLM 5.3 Flash | 321B MoE | Offload Q2_K, 128 GB RAM · ~6.4–12 tok/s |
| Gemma 4 31B IT | 31B | Offload Q4_K_M, 32 GB RAM · ~1.6–3.2 tok/s |
| gpt-oss 20B | 21B MoE | Offload Q4_K_M, 32 GB RAM · ~27–45 tok/s |
| Gemma 4 26B A4B IT | 26B MoE | Offload Q4_K_M, 32 GB RAM · ~24–39 tok/s |
| Qwen3.6 35B A3B | 36B MoE | Offload Q4_K_M, 32 GB RAM · ~30–44 tok/s |
| Qwen3 32B | 33B | Offload Q4_K_M, 32 GB RAM · ~1.6–3.2 tok/s |
| Qwen3.6 27B | 28B | Offload Q4_K_M, 32 GB RAM · ~1.9–3.7 tok/s |
| gpt-oss 120B | 117B MoE | Offload Q4_K_M, 64 GB RAM · ~12–23 tok/s |
| DeepSeek V4 Flash 0731 | 304B MoE | Offload Q2_K, 128 GB RAM · ~2.3–4.7 tok/s |
| Qwen3 Coder 30B A3B Instruct | 31B MoE | Offload Q4_K_M, 32 GB RAM · ~24–44 tok/s |
| Qwen3.5 27B | 28B | Offload Q4_K_M, 32 GB RAM · ~1.9–3.7 tok/s |
| Qwen3.5 35B A3B | 36B MoE | Offload Q4_K_M, 32 GB RAM · ~30–44 tok/s |
| DeepSeek V4 Flash | 291B MoE | Offload Q2_K, 128 GB RAM · ~8.3–15 tok/s |
| GLM 4.5 Air | 110B MoE | Offload Q4_K_M, 128 GB RAM · ~2.5–5.2 tok/s |
| MiniMax M2.5 | 229B MoE | Offload Q3_K_M, 128 GB RAM · ~6.3–12 tok/s |
| Hy3 | 299B MoE | Offload Q2_K, 128 GB RAM · ~4.9–9.5 tok/s |
Comparing cards? The RX 7600 (launch MSRP $269) is in best GPUs for local LLMs under $300. See also Mac vs PC.
Just too big at 4-bit
These need up to 35% more memory than the RX 7600 has at Q4_K_M and 8K context. A smaller quant or shorter context may work.
| Model | Q4_K_M needs | Short by |
|---|---|---|
| Llama 3.2 11B Vision | 8.0 GB | 0.4 GB |
| KORMo Web 11B | 8.3 GB | 0.7 GB |
| DeepSeek Coder 6.7B Instruct | 8.3 GB | 0.7 GB |
| DeepSeek Coder 6.7B Base | 8.3 GB | 0.7 GB |
| DeepSeek Coder 7B Instruct v1.5 | 8.3 GB | 0.7 GB |
| Llama 2 7B HF | 8.3 GB | 0.7 GB |
| Gemma 3 12B IT | 8.5 GB | 0.9 GB |
| Llama Guard 4 12B | 9.1 GB | 1.5 GB |
| DeepSeek V2 Lite | 9.8 GB | 2.2 GB |
Questions
What is the biggest model that runs on the RX 7600?
Of the models we track, the largest that fits at 8K context is Kimi VL A3B Thinking 2506 (16B) at Q2_K, needing about 6.7 GB.
How much of the RX 7600's memory can a model use?
About 7.6 GB: we assume 95% of the 8 GB is usable after driver and display overhead.
Spec source: AMD specifications. Model data: Hugging Face Hub API. Speed is a bandwidth ceiling (bandwidth ÷ bytes read per token), not a benchmark. Methodology · Report an error.