Back to Home
RadarAI··Papers & Tech

Gemini 4 发布即「屠榜」,但我不敢半场开香槟

中文摘要

Gemini 4 Argon 领跑基准测试,但编程与Agent能力弱。官方评价存疑,其实际表现仍待验证。

English Summary

Gemini 4 Argon leads many benchmarks, but programming and agent skills are weak. Official scores dispute internal views; its true performance needs verification.

Original Excerpt

📌 一句话摘要 Gemini 4 Argon 发布即领跑多项基准测试,但编程与 Agent 能力相对落后,加之官方跑分与内部员工评价存在分歧,其实际表现仍需时间验证。 📝 详细摘要 Google Gemini 旗舰模型时隔七个半月更新为 Gemini 4 Argon,在 18 项官方测试中领先 13 项,尤其在知识工作(Vals Finance Agent、Harvey's Legal Agent Benchmark)...