Gemini 4 Argon 基准图：19 项中 13 项登顶
图表显示 Argon 在 19 项结果中 13 项单独领先，长上下文 GraphWalks 256K-1M 达 84.2%，Harvey 法律基准 19.6% 近对手三倍，但 FrontierSWE 与 Terminal-Bench 仍落后 Astra 和 Opus 5.5。
Google has finally released a stable pro model after more than a year and a half of waiting!! Gemini 4 Argon leads outright on 13 of the 19 results on this chart and ties Astra on cybersecurity. The long-context gap is pretty good, (they didn’t report it but they’re still behind Muse 1.3)- 84.2% on GraphWalks at 256K-1M tokens, compared with 71.8% for Astra and 66.8% for Opus 5.5. IMG_2573.jpeg On Harvey’s legal benchmark, it scores 19.6% against 6.7% for the next-best model shown, almost 3x the score. It also leads all four knowledge-work evaluations in this table, including 51.3% on AutomationBench against 42.5% for Opus and 41.4% for Astra. Astra still leads FrontierSWE and Terminal-Bench Science, while Opus 5.5 leads Terminal-Bench 4.0 and PostTrainBench. So the strongest argument for Argon here is the breadth of its results, especially across knowledge work and long-context reasoning.
这条是引用着另一条写的：x.com/googledeepmind/status/2105388084154056939