All hardware / RTX 5090
With 32 GB of VRAM, the RTX 5090 runs 15 of 22 popular open models fully in fast memory at Q4_K_M or better (8K context, assuming 32 GB of DDR5 system RAM). The largest that fit are Hermes 4.3 36B (Q5_K_M), Qwen3.6 35B-A3B (Q6_K), Gemma 4 31B (Q6_K). Fastest: Gemma 4 E2B at over 150 tokens/sec and Qwen3.6 35B-A3B at over 150 tokens/sec. 1 more runs with offload to system RAM, but slower.
Check with your own RAM and context length| Model | Result | Best quant | Memory needed | Est. speed |
|---|---|---|---|---|
| Gemma 4 E2B | Runs on GPU | Q8_0 | 6.1 GB | 150+ tok/s |
| Qwen3.6 35B-A3B | Runs on GPU | Q6_K | 29.7 GB | 150+ tok/s |
| gpt-oss-20b | Runs on GPU | Q8_0 | 23.2 GB | 150+ tok/s |
| Ministral 3 3B | Runs on GPU | Q8_0 | 4.8 GB | 150+ tok/s |
| Gemma 4 26B A4B | Runs on GPU | Q8_0 | 27.9 GB | 150+ tok/s |
| Gemma 4 E4B | Runs on GPU | Q8_0 | 9.3 GB | 150+ tok/s |
| Qwen3.5 9B | Runs on GPU | Q8_0 | 11.3 GB | 104 tok/s |
| Gemma 4 12B | Runs on GPU | Q8_0 | 13.8 GB | 84 tok/s |
| Phi-4 14B | Runs on GPU | Q8_0 | 17.7 GB | 69 tok/s |
| Granite 4.2 30B | Runs on GPU | Q6_K | 24.8 GB | 45 tok/s |
| Gemma 4 31B | Runs on GPU | Q6_K | 26.4 GB | 43 tok/s |
| Mistral Small 3.1 24B | Runs on GPU | Q8_0 | 27.3 GB | 42 tok/s |
| Hermes 4.3 36B | Runs on GPU | Q5_K_M | 27.7 GB | 42 tok/s |
| Qwen3.6 27B | Runs on GPU | Q8_0 | 29.8 GB | 37 tok/s |
| Qwen3.8 27B | Runs on GPU | Q8_0 | 30.6 GB | 36 tok/s |
| Llama 3.3 70B | Needs offload | Q4_K_M | 45.9 GB | 2.9 tok/s |
| Llama 4 Scout | Too big | Q4_K_M | 68.1 GB | — |
| gpt-oss-120b | Too big | Q4_K_M | 72.0 GB | — |
| Qwen3.5 122B-A10B | Too big | Q4_K_M | 75.3 GB | — |
| Qwen3.8-Flash-Next | Too big | Q4_K_M | 107.8 GB | — |
| GLM-5.3-Flash | Too big | Q4_K_M | 196.0 GB | — |
| DeepSeek V4.1-Flash | Too big | Q4_K_M | 335.8 GB | — |
OllamaLM StudioJanGPT4AllAnythingLLMllama.cppvLLMSGLangKoboldCppText Generation WebUILocalAI