Gemini Carbon 曝光：基于 Argon 的 Gemini 4 新成员
传闻 Gemini 4 新模型 Carbon 已向 Google 内部和少数伙伴开放，主打长程编码、自主 agent 与 3D 生成，速度比 Argon 慢 2–4 倍。
Gemini Carbon is a new model within the Gemini 4 lineup, an iterative evolution built upon the released Gemini 4 Argon architecture. Early access is live for Google internal teams and a small group of partners. It’s deployed on Google’s internal code platform Jetski for select staff testing. Deep reasoning: Stronger logical reasoning compared to Gemini 4 Argon.Core focus areasLong-horizon coding: Tuned for complex, extended programming work, closing gaps that held back prior models on lengthy coding tasks.Autonomous agent workflows: Supports multi-step, self-executing agentic workloads.3D content generation: Enhanced 3D scene generation. Speed & stability: It runs 2–4× slower than Argon, but delivers superior stability and sustained reasoning for complex tasks.Performance notesCoding: Internal testers report Carbon’s code output sits near or on par with Anthropic’s Claude Opus 5.5. Whispers say Carbon has achieved an RSI breakthrough. It can autonomously patch code and iteratively improve its own abilities, stoking industry chatter that AGI may be drawing near. It’s positioned to go head-to-head with Opus 5.5 — at drastically better value.
这条是引用着另一条写的：x.com/bindureddy/status/2114906803452690453