AI Arena
Independent AI model benchmarks on real hardware. Every test runs on the same rig — we measure speed, quality and resource usage.
Test rig
CPU
AMD Ryzen
GPU
16 GB VRAM
RAM
32 GB DDR
OS
Ubuntu 24.04 LTS
gemma4:e4b
Google
Text Code
84.0
tok/s
84.8
tok/s
84.4
tok/s
84.3
tok/s
gemma4:12b-q4_K_M
Google
Text Code
42.4
tok/s
42.3
tok/s
42.6
tok/s
42.5
tok/s
qwen3:32b-q4_K_M
Alibaba (Qwen)
Text Code
3.6
tok/s
qwen3-coder:30b
Alibaba (Qwen)
Text Code
9.8
tok/s
deepseek-coder-v2:16b
DeepSeek
Text Code
13.8
tok/s
14.4
tok/s
15.9
tok/s
17.6
tok/s
DeepSeek V4 Pro (cloud)
DeepSeek (cloud)
Text Code
4,096,000.0
tok/s
176.0
tok/s
4,096,000.0
tok/s
614.2
tok/s
Gemini 3.6 Flash (cloud)
Google (cloud)
Text Code
286.3
tok/s
281.5
tok/s
277.6
tok/s
155.6
tok/s
Claude Sonnet 4.5 (cloud)
Anthropic (cloud)
Text Code