Artificial Analysis has introduced AA-AgentPerf, a new benchmark designed to evaluate inference hardware by simulating real-world coding agent behaviors. This benchmark measures total server throughput per megawatt of GPU power against per-request output speed. In tests involving models like DeepSeek V4.1 Flash, DeepSeek V4 Pro 0813, Kimi K3, and GLM-5.2/5.3, NVIDIA's Vera Rubin hardware demonstrated superior performance, leading in energy efficiency across various speeds. AI
IMPACT Sets a new standard for evaluating AI inference hardware, potentially guiding future development towards more energy-efficient solutions.
RANK_REASON New benchmark for AI inference hardware performance and energy efficiency. [lever_c_demoted from research: ic=1 ai=0.7]
- AA-AgentPerf
- Artificial Analysis
- DeepSeek V4.1 Flash
- DeepSeek V4 Pro 0813
- GLM-5.2
- GLM-5.3
- Kimi K3
- NVIDIA
- Vera Rubin
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →