Particle launches Particle-1 with lower coding output token usage

Particle post-trained DeepSeek V4.1 Flash in its own reinforcement learning environment to reduce token usage on software engineering tasks. Across four internal benchmarks, it reports 28.4–52.2% fewer output tokens and 34.4–51.6% lower average task costs, with Pass@1 within the base model's confidence interval.

Visit original source