Particle-1, a token-efficient coding model, arrives on Vercel's AI Gateway

Vercel has added Particle-1, a coding model from Particle AI, to its AI Gateway, giving developers a new option tuned to complete software engineering work while consuming fewer tokens.

A coding model built for efficiency

According to the Vercel blog, Particle-1 is post-trained on DeepSeek V4.1 Flash and is aimed squarely at software engineering tasks. Its pitch is token efficiency rather than headline benchmark scores: Particle describes the model's advantages as shorter reasoning chains, fewer cycles of writing code and then repairing it, and more targeted exploration of a problem.

The model ships with a 1 million-token context window, which Vercel positions as suitable for working with large amounts of code and extended task histories, such as long agent sessions where earlier debugging steps need to remain in context. Suggested use cases include debugging, planning code changes, and breaking down multi-step engineering problems.

How to get started

Particle-1 is reachable through AI Gateway using the model name particle/particle-1, with authentication handled by an AI_GATEWAY_API_KEY environment variable. Because the gateway exposes a unified API, the model can be dropped into existing agent tooling. Vercel says installing the latest version of its CLI and running npx vercel ai-gateway setup configures agents such as Claude Code, Codex and Cursor, after which the user selects particle/particle-1 in their tool of choice. A model playground is also available for trying the model before wiring it into a workflow.

What the gateway adds

AI Gateway itself is Vercel's routing layer for language models: a single API for calling models from different providers, tracking usage and cost, and configuring retries, failover and other performance optimizations, which Vercel says deliver higher uptime than calling providers directly. The service also includes custom reporting, Zero Data Retention support, per-key budgets, and routing rules.

On pricing, Vercel says AI Gateway passes through provider pricing with no markup and no platform fee on inference, including on Bring Your Own Key requests where developers supply their own provider credentials.

Why it matters

Token consumption is a large share of the real cost of AI-assisted coding, and agentic workflows tend to make it worse. Agents that reason at length, or that iterate through failed attempts before landing on a fix, can burn through both context and budget quickly. A model deliberately trained to reason more briefly and repair less attacks that cost curve directly, and the 1M-token context window means longer sessions are less likely to force painful truncation.

For developers already routing models through Vercel, particularly those using AI Gateway as the backbone for tools like Claude Code or Cursor, Particle-1 is a low-friction experiment: one CLI command, no separate account, and provider-level pricing. If its efficiency claims hold up in everyday engineering work, it becomes a credible choice for routine debugging and planning tasks. If they do not, the gateway's failover and routing rules make it straightforward to fall back to another model without rewriting application code.