DeepSeek V4 Flash beats Qwen3.8-27B in SparkBench evaluation

solyarisoftware · x · 2026-08-19

SparkBench results show DeepSeek V4 Flash scoring 93.01, defeating Qwen3.8-27B's 90.94. DeepSeek wins in code, agents, and tools, while Qwen leads in robustness and calibration. DeepSeek V4 Flash also demonstrates lower median latency and is less verbose.

Original post →

More from Models

Models channel →