"Use independent third-party benchmarks to prove quality Because quality is "always the hard one" to measure credibly, Inception gave their model to Artificial Analysis — an independent evaluator — and let them run their own benchmark suite without Inception controlling the process. The independent score confirmed Mercury 2 is "comparable in quality to the speed-optimized models from Frontier Labs" while being "5–10x faster." Whenever you are making a quality claim about your own work, handing the evaluation to a credible independent party is far more persuasive than self-reported numbers, whether in AI, product development, or research. Source: Arash Vahdat, Latent Space Podcast, Diffusion LLMs with Inception AI"