谷歌 Gemini 4 Argon 逼近 AI 前沿，但测试方法与数学编码仍存短板
原标题：Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months
AI 摘要
Google DeepMind 发布新前沿模型 Gemini 4 Argon，在 ARC-AGI-3 基准上以 97.5% 的成绩成为继 OpenAI 之后第二家突破 95% 的实验室，仅次于 GPT-6 Astra 的 99.9%。但该成绩依赖使用厂商特定功能的 Provider Adapter Harness 测试环境，在仅允许简单笔记的 Standard Harness 下低推理档仅得 49.8%。Epoch AI 的 ECI 综合指数显示 Argon 得 165 分，与 Fable 5.1 持平，落后于 Opus 5.5 和 GPT-6 Astra，且在数学和软件工程领域仅排第八和第六。
正文节选
Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months Google Deepmind is the second major AI lab to pass 95 percent on the ARC-AGI-3 benchmark. But the results depend heavily on the testing method, and Argon trails rivals in math and software engineering, according to Epoch AI. Google Deepmind's new frontier model, Gemini 4 Argon, scored 97.5 percent on ARC-AGI-3 at a cost of $4,800 using the "Provider Adapter Harness," a test setup that uses vendor-specific featur