How-to

Gemini 4 三个版本选用指南:Neon、Hydrogen、Carbon 怎么分工

Why we picked it

Phil Schmid 把 Gemini 4 三个版本的使用分工讲得很清楚:Neon 当编码主力,Hydrogen 跑子任务省一半钱,Carbon 留给慢速模拟。正在搭多智能体工作流、想压 token 成本的人可以直接抄这套路由方案。

Phil Schmid 分享了 Gemini 4 三个版本的模型选型建议。Neon 定价为 $3/$15(输入/输出每百万 token),被他作为编程和智能体任务的日常主力模型。Hydrogen 定价为 $1.50/$7.50,速度最快、成本最低,适合基础任务、聊天或子智能体调用。Carbon 则用于可以接受更慢响应的场景,如测试和复杂长周期任务的模拟运行。

Body

这篇指南的核心是把模型按成本和延迟分层:高频的编码与智能体工作负载交给 Neon,批量或简单任务用 Hydrogen 降本,长时程模拟类任务才启用 Carbon。价格上 Hydrogen 比 Neon 便宜一半,适合作为子智能体的默认调用模型来压低多智能体系统的总成本。

对于搭建智能体工作流的开发者来说,这种按任务类型分配模型的思路可以直接套用到自己的路由策略里,用最便宜的模型处理子任务、保留较强模型处理主线逻辑。

Original · Philipp Schmid

My Gemini 4 model guidance: Neon = Daily driver for coding and agents ($3/$15). I mostly use Neon now. Hydrogen = fastest, cheapest ($1.50/$7.50). Use for basic tasks/chat or subagents. Carbon = Opened tasks that are ok to be slower. Testing, Simulations for complex long-horizon runs📊 16 ⚡

Industry Coordinates & Source Verification

This report is categorized under AITOP's "How-to" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Gemini 4, Neon, Hydrogen, prompt engineering.

Source attribution: curated from Philipp Schmid (@_philschmid). Cross-source corroboration: 2 independent sources have covered this development (including lmarena.ai), confirming industry-wide signal strength.