随着谷歌准备向公众推出其全新的 Gemini 4 “Argon” 模型，公司内部员工正在测试性能更优的版本，这些版本或许能进一步缩小与竞争对手的差距。根据《商业内幕》（Business Insider）看到的文件和截图，最近，员工们正在内部试用一个名为 Carbon 的 Gemini 4 新版本。该模型可能是 Argon 的又一次更新，也被称为“检查点”（checkpoint），近日已在谷歌内部的编码平台 Jetski 上开放使用。
As Google prepares to roll out its new Gemini 4 'Argon' model to the public, staff inside the company are testing better versions that may further close the gap with rivals. Most recently, employees have been trying a new version of Gemini 4 internally named Carbon, according to documents and screenshots seen by Business Insider. The model — possibly a new update to Argon, also known as a "checkpoint" — was made available inside Google's internal coding platform, Jetski, in recent days.
一名员工告诉《商业内幕》，在编码方面，Carbon “感觉就像 Opus 5.5”，指的是 Anthropic 专为长期自主编码设计的最强模型。不过，该员工也补充说，他们还需要对它进行更多测试。谷歌的内部测试表明，该公司正在快速推进其于 9 月底宣布的新一代 Gemini 4 模型。这也凸显了谷歌在激烈竞争的 AI 编码代理赛道上的努力。
One employee told Business Insider that Carbon "feels like Opus 5.5" for coding, referring to Anthropic's most powerful model for long-term agentic coding. However, they caveated that they would need to test it more. Google's internal testing shows how the company is moving fast with its new generation of Gemini 4 models, which it announced in late September. It also underscores how Google is competing in the fast-moving race for AI coding agents.
Anthropic 和 OpenAI 凭借其专为复杂工程任务设计的模型，在开发者群体中已处于领先地位。根据《商业内幕》与员工的交流以及内部消息记录，Argon 的早期内部测试版本总体上获得了员工的认可。不过，一名员工表示，Argon 的早期版本在某些编码任务上显得落后，并将其与 Anthropic 较旧的 Claude Opus 5 模型进行了比较。
Anthropic and OpenAI have pulled ahead among developers with models built for complex engineering work. Early versions of Argon tested internally were generally well received by staff, according to conversations with employees and internal message threads seen by Business Insider. However, one employee said early versions of Argon felt behind on some coding tasks, comparing the model to Claude Opus 5, an older Anthropic release.
最新的更新或许能弥补这一差距，但谷歌并不一定会发布这些版本。一份内部文件显示，谷歌已测试了一系列内部命名为 Argon、Barium 和 Carbon 的 Gemini 4 模型。谷歌的内部代号通常与其公开名称并不一致。例如，一份文件指出，内部名为 “Barium-B” 的模型最终被选为将以 Argon 之名面向公众的版本。目前尚无法确定 Carbon 是会作为 Argon 的又一次更新发布，还是 Gemini 4 系列中的另一个独立模型。谷歌拒绝对此发表评论。
The latest updates might close that gap, but Google is not guaranteed to release them. An internal document shows that Google has tested a series of Gemini 4 models, internally named Argon, Barium, and Carbon. Google's internal codenames are typically out of sync with their public names. For example, one document noted that the internal "Barium-B" model was chosen to be the model that will be known publicly as Argon. It could not be determined whether Carbon will be another update under the Argon name or a different model in the Gemini 4 family. Google declined to comment.
员工们一直在消息渠道中讨论这些新更新，Business Insider 也看到了这些讨论。一名员工表示：“感觉和 Opus 5.5 差不多。”他指的是最新的更新。另一位员工写道：“Carbon 真的很好！”还有人发帖称：“新的 Gemini Pro Next 模型感觉还挺不错的。”他指的是公司内部通常用于新 Gemini 更新的“Next”标签。谷歌上个月表示，Argon 在某些编码和知识工作基准测试中实现了前沿性能，涵盖法律和金融等领域。
Staff have been discussing the new updates in messaging channels seen by Business Insider. "Feels comparable to Opus 5.5," said one employee, referring to the latest update. "Carbon is really good!" another wrote. "New Gemini pro next model feels kinda nice," another person posted, referring to the "Next" label that is typically given to new Gemini updates internally. Google said last month that Argon delivered frontier performance on some coding and knowledge work benchmarks, including domains such as legal and finance.
该公司还表示，Argon 特别注重防御性网络安全。谷歌称，在更广泛地面向公众推出之前，会首先通过其 Fairwind 计划向合作伙伴提供该模型。该计划旨在测试新模型中的网络漏洞。谷歌正面临压力，需要证明其模型能力在前沿领域仍然具有竞争力。本周，该公司宣布推出了一款专为工作场所设计的新“通用” Gemini AI 代理，而各大 AI 公司之间围绕智能代理助理的竞争也愈发激烈。
The company also said Argon had a strong focus on defensive cybersecurity. It said it would first make the model available to partners in its Fairwind Program, an initiative for testing cyber vulnerabilities in new models, before a wider public rollout. Google is under pressure to show it can still compete on the frontier in terms of model capabilities. The company announced this week a new "universal" Gemini AI agent designed for the workplace, as the battle over agentic assistants among top AI companies heats up.