Loading…
Get the best builds by email. No account needed.
Live drop
DiffuSpaceDropped Oct 30, 2026
Block-diffusion MoE LLM enables faster agentic coding and tool-use generation than same-size autoregressive models.
About this release
DiffuSpace's first open-weights model: a 35B-total / ~3B-active MoE diffusion language model with block diffusion generation, thinking mode, and 128K context. Built on a Qwen-derived backbone, it supports bidirectional attention within blocks for parallel denoising while preserving causal prefix. Released under Apache 2.0 with Hugging Face, SGLang, and llama.cpp inference paths.
Read the official DiffuSpace launch pageFirst builds incoming — this lane fills in automatically as people start building with Clover-35B-A3B-Preview. Know one? Submit the URL.