China's Floatboat Harness Beats Claude Opus 4.8 on All Five Benchmarks While Running on the Cheapest Model on the Market
· pandaily.com
AOE Tech Labs' Floatboat team ran a single-variable Harness benchmark experiment on August 7. With DeepSeek-V4-Flash as the model base and only the execution Harness swapped, the same 0.14-dollar model went from losing all five benchmarks against Opus 4.8 on DeepSeek's own Harness to winning all five on Floatboat's Harness, at 57.1 times lower cost.
AOE Tech Labs' Floatboat team ran a single-variable Harness benchmark experiment on August 7. With DeepSeek-V4-Flash as the model base and only the execution Harness swapped, the same 0.14-dollar model went from losing all five benchmarks against Opus 4.8 on DeepSeek's own Harness to winning all five on Floatboat's Harness, at 57.1 times lower cost.