Unverified90% confidenceFactTime unknown
Mercury 2 采用扩散模型方式，先并行生成整个回答的初稿，然后通过多轮迭代逐步润色优化，而非自回归模型的逐词生成方式
1
Sources
90%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedInception Labs released Mercury 2, a large language model built on a diffusion model architecture, claimed to be the fastest-reasoning model in the world.78% similarVerifiedThe Mercury series represents a commercial breakthrough for discrete diffusion language models transitioning from academic prototypes to production-ready deployment.72% similarUnverifiedThe technical foundations of diffusion models originate from stochastic processes in non-equilibrium thermodynamics, comprising two stages: a forward diffusion process and a reverse denoising process.61% similarVerified扩散模型是一类基于马尔可夫链的生成模型，核心思想是通过前向加噪和反向去噪来捕捉数据分布60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/14049API
curl https://kongchang.com/api/v1/knowledge/claims/14049MCP
get_claim(id=14049)