Home/New releases/Particle-1

Particle-1

A coding model from Suhail Doshi's new company, post-trained on DeepSeek V4.1 Flash to use fewer tokens per task.

Particle announced Particle-1 on October 28, 2026. It is post-trained on DeepSeek V4.1 Flash with reinforcement learning to finish tasks with fewer tokens. The first release focuses on software engineering. Particle's numbers: it uses about 45% fewer tokens on average, and 28.4% to 52.2% fewer output tokens on its four coding benchmarks, with pass rates within the base model's confidence intervals. Particle's blog footnote lists the price per million tokens as $0.30 uncached input, $0.006 cached input, and $1.20 output, the same as DeepSeek V4.1 Flash. Context is 1,048,576 tokens. The API is OpenAI-compatible and supports tool calling. It is available through the Particle API and Vercel AI Gateway. No weights release is stated. Last checked October 29, 2026.