Particle-1

Particle-1 is Particle AI's first model, released 2026-10-28: a token-efficient agentic coding and software-engineering model post-trained on DeepSeek V4.1 Flash to finish tasks (terminal workflows, codebase changes) with fewer tokens. Text and image input, text output; reasoning, function calling and JSON-schema structured outputs; 1,048,576-token context, 131,072 max output. Served via the Particle API, OpenRouter (particle-ai/particle-1) and Vercel AI Gateway; the official launch post lists $0.30 / $1.20 per 1M uncached input/output tokens ($0.006 cached input). Particle reports ~45% fewer tokens on average vs. base DeepSeek V4.1 Flash at comparable accuracy. Proprietary; weights not released. No vendor benchmark table mapped at launch.