LectureRechercheRadarCadre d'investissement
Connexion / Inscription
← Retour à la lecture
ActualitésSujet · Particle-1模型Prise rapide1 min de lectureRédaction DisconfirmAI

Vercel AI Gateway Adds Particle-1 for Efficient Software Engineering Tasks

Vercel AI Gateway now supports Particle-1, a model post-trained on DeepSeek V4.1 Flash that claims to complete coding tasks with fewer tokens.

Vérifié 2026-10-29 06:05 GMT+8

Vercel has announced that its AI Gateway now supports Particle-1, a model from Particle AI. Post-trained on DeepSeek V4.1 Flash, the model is designed to complete software engineering tasks such as debugging, planning code changes, and breaking down multi-step problems using fewer tokens.

Particle-1 features a 1 million-token context window, suitable for working with large codebases and extended task histories. Developers can access it by setting the AI_GATEWAY_API_KEY environment variable and using particle/particle-1 as the model name. It is also compatible with coding agents like Claude Code, Codex, and Cursor via the latest Vercel CLI setup.

Vercel states that AI Gateway provides a unified API for calling models, tracking usage, and configuring retries without charging a platform fee or markup on inference. The claimed gains in shorter reasoning and reduced trial-and-error rounds are based on vendor descriptions, with no independent benchmark results provided.

Sources

  • vercel.comVercel↗
Sujet

Particle-1模型

Particle AI于2026年10月28日发布了其首个模型Particle-1,该模型基于DeepSeek V4.1 Flash进行后训练,旨在通过强化学习优化软件工程任务中的Token使用效率。【背景】Particle由Mixpane…

Connexion / Inscription
LectureRechercheRadarCadre d'investissement

3 étapes1 piècesMis à jour 2026-10-30 08:02

Comment on en est arrivé là

  1. 2026-10-28Particle AI正式推出Particle-1模型,宣称基于DeepSeek V4.1 Flash后训练,专注于软件工程任务的Token效率优化。particle.ai ↗
  2. 2026-10-29Vercel宣布AI Gateway新增对Particle-1的支持,允许开发者通过统一API调用该模型进行调试和代码规划。vercel.com ↗
  3. 2026-10-29OpenRouter上架Particle-1,定价为输入$0.25/百万Token,输出$1.00/百万Token,高于DeepSeek原生价格。cellcog.ai ↗

Ce qui reste non résolu

  • 是否有独立第三方机构复现了Particle-1在Terminal-Bench 2.1和DeepSWE 1.1上的Token节省数据?
  • Particle-1的LoRA适配器权重是否会开源,以便社区自行部署或进一步微调?
  • 在非编码类通用任务(如Tau Airlines)上,Particle-1相对于基座模型的细微性能差异是否会在复杂Agent工作流中被放大?
Lire le sujet complet →
Continuer la recherche →