Particle-1 lands on AI Gateway: a software-engineering model built to finish coding tasks with fewer tokens
Synopsis
Particle AI's Particle-1 is now available on AI Gateway, giving developers a unified endpoint for the model. Particle-1 is post-trained on DeepSeek V4.1 Flash, and Particle describes its gains as shorter reasoning, fewer rounds of trial and repair, and more focused exploration, aimed at debugging, planning code changes and breaking down multi-step engineering problems. It offers a 1M-token context window for working with code and extended task histories.
Interpretation
Developers can now call Particle AI's Particle-1 through AI Gateway and use it for debugging, planning code changes and breaking down multi-step engineering problems. Particle frames Particle-1's improvement as completing software engineering tasks with fewer tokens rather than adding new task types. Particle describes the gains as shorter reasoning, fewer rounds of trial and repair, and more focused exploration; these are vendor descriptions of benefits, with no comparison baseline or quantitative result attached.
Particle-1 is post-trained on DeepSeek V4.1 Flash, so its capability comes from continued training on an existing model rather than a new base model trained from scratch. Post-training means continuing training on an existing base model to adjust its behavior, and Particle chose DeepSeek V4.1 Flash as that base. The model offers a 1M-token context window for working with code and extended task histories, a specification that matches debugging and multi-step use.
Access follows AI Gateway's unified calling path: developers set AI_GATEWAY_API_KEY and request the model name particle/particle-1. AI Gateway itself provides a unified API for calling models, tracking usage and cost, and configuring retries, failover and performance optimizations. To use it in agent tools such as Claude Code, Codex and Cursor, users install the latest Vercel CLI and run npx vercel ai-gateway setup, then select particle/particle-1 in the agent.
Perspective
The report covers Particle-1's availability on AI Gateway, its post-training base, Particle's qualitative description of the gains, the 1M-token context window and the access path; the intended reader is a developer who needs to call the model inside agent tools.
Particle's description of the gains is qualitative, with no comparison baseline, task set or quantitative metric; how much difference fewer tokens and fewer repair rounds make on specific engineering tasks remains to be observed in use.