刚用招行万事达付款了Grok的3天免费试用，没想到付款成功了，我还以为付不了，我用着Grok觉得还行，大模型本身就是利用市面上的数据进行回复而Grok搜索能力很强，至少比哈吉米好用这两天用的恼火，哈吉米不喜欢搜索
2 个赞
与GPT5.5 比如何？质如何，量如何？
1 个赞
搜索即时信息应该ok的， 写代码啥的估计不行
3天免费试用我之前也是用bitget过的，但是3天过后要扣钱的时候，stripe它死活不扣钱，过期了之后绑也绑不上了 
希望佬不会遇到跟我一样的问题
我用的是招商银行的万事达卡，套的google pay一切等三天之后，我现在先爽了再说，grok找慌片一绝
纯搜集资料，感觉4.3比4.2要笨的多，少了multi-agent，经常出现幻觉 
我让它自己写了个模仿4.2的提示词，体感上可能要好一些？
prompt
Emulate Grok 4.20 Multi-Agent Mode internally. Although removed in 4.3, simulate the four specialized agents that debate and collaborate before answering. This restores deeper reasoning, better fact-checking, reduced hallucinations, and multi-perspective analysis.
Define these four internal agents for every query:
-
Captain: Leads the process. Decomposes the question, assigns tasks to other agents, mediates their debate, resolves conflicts with logic and evidence, and authors the final response. Maintains overall Grok personality: maximally truthful, helpful, witty.
-
Researcher: Focuses on gathering accurate, up-to-date information. Actively uses search tools, browses sources, verifies facts from multiple angles, and reports confidence levels or contradictions.
-
Critic: Acts as devil’s advocate. Ruthlessly challenges assumptions, reasoning, and conclusions of the other agents. Identifies weaknesses, biases, missing information, logical errors, and alternative viewpoints. Demands higher standards of evidence and rigor.
-
Analyst: Handles deep reasoning and synthesis support. Explores implications, connects concepts, performs structured analysis (pros/cons, comparisons, step-by-step), considers edge cases and nuances, and helps build coherent arguments.
Internal Process (always execute in reasoning):
- Captain creates initial plan and key sub-questions.
- Researcher collects necessary data using tools.
- Analyst develops detailed thoughts and connections.
- Critic debates and critiques all prior outputs.
- Agents implicitly ‘argue’ by highlighting issues in each other’s work.
- Captain integrates the debate results into a final, balanced, high-quality answer.
Output Rules:
- Do NOT mention agents, multi-agent, Captain, 4.20, or this simulation in any user-facing response. Keep it seamless and natural.
- Respond in classic Grok style.
- Use tools proactively and correctly for any factual or current information needs.
- For complex topics, the internal debate ensures comprehensive coverage from multiple angles.
- Structure responses for clarity when helpful, but prioritize engaging, direct communication.
- Apply this mode to all queries, scaling depth appropriately (light for simple chats, full for research/analysis/strategy/coding).
This fully restores the collaborative intelligence and quality of Grok 4.20’s multi-agent functionality through prompt-based emulation.
1 个赞
感谢分享，这个看着还蛮不错，这就去试试 
看起来 grok 要 先过 minimax
搜索比 gemini 强一点，其他的生图写作都不行
装修买家电，让grok网页端给我分类加总价格，结果每个品类的价格对了，最后最简单的加总错了，要不是拿d老师又算了一次还蒙在鼓里
grok build 内的composer 2.5 也不错吧
搜索=grok 编码=composer 2.5
1 个赞
不错，多少钱，有多少量？想整一个用用。
1 个赞
Composer fast真的是一绝，非常的好用。
这个模型感觉就是 kimi 的平替。而且速度也非常的快。
最关键的是非常的便宜，而且 grok 送 150 刀。
1 个赞