版本亮点
- Google 介绍 Carbon 在 Gemini 4 Argon 基础上研发,重点面向长流程工程与网络安全防御。 Google 发布介绍
相关动态
订阅更新公开成绩与计算依据
综合参考分根据已公开的评测资料整理。查看发布日期依据
综合参考分 93.9 · 13项基准
推理求解 98.7
Terminal-Bench-Science · 0.647619047619048 比例
Artificial Analysis · 独立评测 · 0.1.0
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; terminal-bench-science
HLE · 0.691 比例
Artificial Analysis · 独立评测 · Diamond
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; hle-diamond
CritPt · 0.69 比例
Artificial Analysis · 独立评测 · 2 Hard
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; critpt-v2-hard
工具与任务执行 90.2
GDPval-AA · 1732.68 分
Artificial Analysis · 独立评测 · 2.1
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; gdpval-aa-normalized
Terminal-Bench · 63.13%
Vals AI · 独立评测 · 4.0
Gemini 4 Carbon / Mini-SWE-agent · Current displayed evaluator row. Mini-SWE-agent. Scores are not pooled across harnesses; source methodology and repeat counts retained.
AutomationBench · 0.7781660572624064 比例
Artificial Analysis · 独立评测 · current published task suite
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; automationbench-aa
AA-Briefcase · 1610.63 分
Artificial Analysis · 独立评测 · 1.1
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; aa-briefcase
OSWorld · 81.5%
blog.google · 官方发布比较 · 2.1
Gemini 4 Carbon / official Gemini 4 Carbon comparison · Official comparison image; OSWorld 2.1 offline subset; original image and release footnotes preserved. Offline-subset and AA-labelled protocols remain distinct. · 2026-10-28公开
编程 89.8
FrontierCode · 0.405606060606061 比例
Artificial Analysis · 独立评测 · 2.0
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; frontiercode-2-0
语言与信息处理 93.7
GDP.pdf · 0.284 比例
Artificial Analysis · 独立评测 · current published task suite
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; gdp-pdf
AA-Omniscience · 0.5826666666666667 比例
Artificial Analysis · 独立评测 · 1.1
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; omniscience-accuracy
AA-Omniscience · 0.8857827476038339 比例
Artificial Analysis · 独立评测 · 1.1
Gemini 4 Carbon (High) · Artificial Analysis current published protocol; omniscience-non-hallucination
多模态理解 100.0
MMMU-Pro · 0.892485549132948 比例
Artificial Analysis · 独立评测 · Standard 10
Gemini 4 Carbon (Medium) · Artificial Analysis current published protocol; mmmu-pro
Chartography · 72.5%
blog.google · 官方发布比较 · current published task suite
Gemini 4 Carbon / official Gemini 4 Carbon comparison · Official comparison image; Chartography ; original image and release footnotes preserved. Offline-subset and AA-labelled protocols remain distinct. · 2026-10-28公开