Particle-1
Particle · 2026-10-28
Particle-1 is Particle's October 2026 agentic coding model, a reinforcement-learning post-train of DeepSeek V4.1 Flash through a rank-64 LoRA adapter, trained to reach solutions with fewer reasoning and tool-call tokens. It is served only through Particle's API (request alias particle-1) and third-party gateways, accepts text and images, generates text, and supports selectable reasoning effort with high as the default.
Benchmark scores
| Benchmark | Score |
|---|---|
| ReactBench (39 tasks) | 33.3 |
| ReactBench (39 tasks): Fixing React (11 tasks) | 14.5 |
| ReactBench (39 tasks): Writing React (28 tasks) | 42.9 |