Google 发布 Gemini 4 Argon：1M 输出、DeepSWE 77.9%
Google 推出 Gemini 4 Argon，先向 Fairwind 网络安全方开放，定价 $2/$10，输出上限从 64K 提到 1M，并给出 DeepSWE、AutomationBench、LVBench 等成绩。
📢 𝗝𝗨𝗦𝗧 𝗜𝗡: $GOOGL Google Unveils Gemini 4 Argon, Its New Frontier AI Model 👉 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: ➤ 𝗚𝗼𝗼𝗴𝗹𝗲 introduces 𝗚𝗲𝗺𝗶𝗻𝗶 𝟰 𝗔𝗿𝗴𝗼𝗻 for complex, long-horizon AI workflows. ➤ Argon initially rolls out to trusted cyber defenders through 𝗙𝗮𝗶𝗿𝘄𝗶𝗻𝗱. ➤ Pricing starts at $𝟮 input and $𝟭𝟬 output per million tokens. ➤ Output limit expands from 𝟲𝟰𝗞 to 𝟭𝗠 tokens for deeper reasoning. ➤ Argon scores 𝟳𝟳.𝟵% on long-horizon software benchmark 𝗗𝗲𝗲𝗽𝗦𝗪𝗘 𝘃𝟭.𝟭. ➤ Model ranks #𝟭 on AutomationBench with a 𝟱𝟭.𝟯% score. ➤ Argon achieves 𝟵𝟭.𝟳% on long-video understanding benchmark 𝗟𝗩𝗕𝗲𝗻𝗰𝗵. ➤ Cybersecurity performance reaches 𝟲𝟴% on 𝗖𝗪𝗘-𝗯𝗲𝗻𝗰𝗵 𝘃𝟭. ➤ Internal optimizations freed over 𝟯𝟬𝟬 𝗧𝗶𝗕 of Google data-center memory. ➤ Broader rollout will begin with paid API and 𝗚𝗼𝗼𝗴𝗹𝗲 𝗔𝗜 𝗨𝗹𝘁𝗿𝗮 customers. 👉 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀: ➤ The 𝟭𝗠-token output limit could enable substantially longer autonomous AI workflows. ➤ Strong coding results target increasingly complex 𝗔𝗜-𝗱𝗿𝗶𝘃𝗲𝗻 software engineering tasks. ➤ Cyber capabilities expand AI's potential role in vulnerability discovery and remediation. ➤ Low introductory pricing could intensify competition across frontier enterprise AI models. ➤ Phased deployment highlights growing emphasis on safeguards for increasingly capable AI agents.
这条是引用着另一条写的：x.com/aistocksavvy/status/2105388980875165801