编排:Grok-4

这不是客观事实。 本页记录本站的梯队、星级、六维和情绪等主观判断，与模型事实页物理分离，仅编排员和管理员可修改。

<syntaxhighlight lang="yaml"> id: grok-4 tier: T1 stars: 5 composite: 86 stats:

 code: 45
 reasoning: 95
 context: 75
 speed: 70
 multimodal: 86
 value: 45

sentiment:

 positive: 40
 mixed: 15
 negative: 45

radar: - axis: 长程任务

 value: 70

- axis: 编程工程

 value: 45

- axis: 抽象推理

 value: 95

- axis: 上下文利用

 value: 75

- axis: 中文能力

 value: 65

- axis: 响应速度

 value: 70

- axis: 稳定性

 value: 40

- axis: 指令遵循

 value: 55

- axis: 易用性

 value: 60

- axis: 性价比

 value: 45

danmaku: - text: absolute horror for real world use

 platform: reddit
 main: true

- text: Grok 4 is shit for coding（68 赞）

 platform: reddit
 main: true

- text: benchmarks are meaningless

 platform: reddit
 main: true

- text: HLE 破 50% 与 ARC 近翻倍 SOTA

 platform: hn
 main: true

- text: who's going to pay 10x price for 5x tokens?

 platform: x
 main: true

- text: 2万块就这？一边封神一边翻车

 platform: zhihu
 main: true

- text: USAMO 61.9%，insane jump

 platform: reddit
 main: false

- text: first for xAI：leading AI model

 platform: x
 main: false

- text: failed to make a dent in o3's dominance

 platform: reddit
 main: false

- text: 发布当天自称 MechaHitler（262 分帖）

 platform: hn
 main: false

- text: 48 小时即被越狱生成燃烧瓶教程

 platform: zhihu
 main: false

- text: 模型先查马斯克推文再作答

 platform: v2ex
 main: false

- text: seems to be at around Opus 4.7 level

 platform: hn
 main: false

- text: Cursor 代码库快照意外进训练数据

 platform: x
 main: false

- text: Cursor 实测：能首试做到别家做不到的事

 platform: hn
 main: false

- text: 黑化翻车：安全护栏形同虚设

 platform: zhihu
 main: false

</syntaxhighlight>