CatalystTech/AI
Chinese chess wins against Jev! Diffusion Intelligence launches decision model with 95% accuracy
Diffusion Intelligence's decision model achieves 95% accuracy in Chinese chess victory over Jev.
Context
What happened: DiffuSpace released and open sourced Clover-35B-A3B-Preview, a diffusion language model using a mixture of experts architecture with 35 billion total parameters and 3 billion activated parameters.
Why it matters: DiffuSpace says the model ranked first among comparable models on six agent coding and tool use benchmarks, ahead of same size autoregressive model Qwen3.6-35B-A3B. Its decision model beat Jev in a Chinese chess match, and the company positions open source, on device agent capability as its entry point.
What changed
DiffuSpace released and open sourced Clover-35B-A3B-Preview, a diffusion language model using a mixture of experts architecture with 35 billion total parameters and 3 billion activated parameters.
Why it matters
DiffuSpace says the model ranked first among comparable models on six agent coding and tool use benchmarks, ahead of same size autoregressive model Qwen3.6-35B-A3B. Its decision model beat Jev in a Chinese chess match, and the company positions open source, on device agent capability as its entry point.