Google 发布 Gemini 4 Carbon，称编码与智能体任务显著提升
Google announces Gemini 4 Carbon with ‘significant gains across coding and agentic tasks’
Google 在 Argon 发布两周后宣布下一代前沿模型 Gemini 4 Carbon，称其在编码、智能体任务和网络安全防御上有显著提升。Carbon 在更新后的 Artificial Analysis Intelligence Index v5.0 上以 59 分排名第一，在 WebDev Arena 上升 89 elo 至 1769 分排名第四。安全方面，Google 称 Carbon 将无依据论断较 Argon 减少约 2 倍，过度越界行为减少 26%、服务不足减少 29%。该模型将先于本周向网络安全防御者开放，随后面向开发者、企业和消费者推出。
推荐理由
精选AI 评分 78
原文列出 Carbon 在编码、智能体任务和安全对齐上的多项基准与内部测试变化，可据此判断这一代模型的侧重点。
主题
正文
Following the launch of Argon two weeks ago, Google today announced Gemini 4 Carbon as its next frontier model.
There are “significant gains” in coding, agentic tasks, and cybersecurity defense (finding vulnerabilities). Notably, “Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59.”
Related: Google adding more Connected Apps to Gemini & Search with new dev platform
Google says this fast release is due to its researchers and models “working hand in hand to advance capabilities faster.”
“With improved quality of output and artifacts in web development, Carbon jumps +89 elo points to land at #4 (1769) on WebDev Arena.”
For cybersecurity:
“firm lead in SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for exploitation, and CWE-bench (71.0%) for patching.”
“On internal benchmarks, Carbon drives a strong uplift in generating secure code, and in live coding environments with our own teams it shows reduced misalignment and fabrication rates.”
On the safety front, Google worked to improve alignment and trained Carbon to “respond to complex or high-pressure tasks in the right way.”
Greater transparency and honesty: In our internal testing, Carbon cuts unsupported claims by approximately 2x over Argon. It puts methodological caveats directly into the deliverable or withholds blocked actions with a clear list of what remains required, rather than fabricating or hedging around missing evidence.
Improved collaboration with the user in the face of ambiguity: Rather than guessing or forcing a path forward, Carbon retrieves the relevant sources and surfaces discrepancies to the user before acting — as a result, users should observe a much better collaborative experience with Carbon.
Following through without overstepping: Carbon simultaneously reduced overreach by 26% and under-service by 29% resulting in much more reliable execution of multi-step plans end-to-end while respecting project constraints and staying strictly within scope.
Officially, Gemini 4 Carbon is “coming soon to developers, businesses, and consumers.” The model will first be available to “cyber defenders” this week.
Updating…