Mercury 2, a diffusion LLM, outperforms StepFun 3.5 Flash on OpenClaw tasks

So Mercury 2 scores significantly higher on speed (96 vs 62), while being comparable on cost efficiency (99 vs 94), consistency (90 vs 89), and raw task success (78 vs 85).

Available via OpenRouter (openrouter/inception/mercury-2) and the Inception API (https://platform.inceptionlabs.ai). Full write-up: https://www.inceptionlabs.ai/blog/mercury-2-on-pinchbench