DefiledAI Community

BENCHMARK SUBMISSIONS

Submit your real-world inference results. Results are stored locally in your browser and displayed below.

Submit Your Result

Model *

Quantization *

Backend *

GPU *

VRAM (GB)

Tok/s (sustained) *

Context Length

Notes

Community Results (10)

Model	Quant	Backend	GPU	VRAM	Tok/s	Ctx	OS	Date
Llama 3.1 8B	Q4_K_M	ExLlamaV2	RTX 4090	24GB	128	4,096	Windows 11	2026-05-27
Llama 3.1 8B	Q8_0	Ollama	RTX 4090	24GB	94	4,096	Windows 11	2026-05-27
Mistral 7B	Q8_0	Ollama	RTX 3080 10GB	10GB	89	4,096	Windows 11	2026-05-23
Phi-3 Medium 14B	Q6_K	llama.cpp	RTX 4090	24GB	68	8,192	Ubuntu 22.04	2026-05-22
Gemma 2 27B	Q4_K_M	llama.cpp	RTX 4090	24GB	44	4,096	Windows 11	2026-05-24
Mixtral 8x22B	Q4_K_M	ExLlamaV2	2× RTX 3090 NVLink	48GB	24.7	4,096	Ubuntu 22.04	2026-05-25
Llama 3.1 70B	Q4_K_M	ExLlamaV2	2× RTX 3090 NVLink	48GB	21.3	4,096	Ubuntu 22.04	2026-05-28
Qwen 3 72B	Q4_K_M	ExLlamaV2	2× RTX 3090 NVLink	48GB	19.8	4,096	Ubuntu 22.04	2026-05-26
DeepSeek R1 70B	Q4_K_M	ExLlamaV2	2× RTX 3090 NVLink	48GB	19.2	4,096	Ubuntu 22.04	2026-05-26
Llama 3.1 70B	Q4_K_M	llama.cpp	2× RTX 3090 NVLink	48GB	17.8	4,096	Ubuntu 22.04	2026-05-28

Your submissions are stored in your browser. Tok/s = sustained output tokens per second, first token excluded.