# Claude-3-Opus 供应商性价比大盘点（上下文窗口200k，不是缩水版的迷你Opus） **URL:** \ **Category:** 资源荟萃\ **Tags:** 人工智能\ **Created:** [2024 年4 月 26 日 08:30 UTC](https://linux.do/t/topic/67372 "2024-04-26T08:30:55Z")\ **Posts on this page:** 20\ **Page:** 2
作者： ![davy](https://cdn.ldstatic.com/user_avatar/linux.do/davy/32/325116_2.png) [@davy](https://linux.do/u/davy)\ 发布日期： [2024 年4 月 26 日 10:39 UTC](https://linux.do/t/topic/67372/22 "2024-04-26T10:39:17Z")
还有OpenRouter上的官方API，输入$15，输出$75，单位都是100万Tokens。 [Anthropic: Claude 3 Opus by anthropic | OpenRouter](https://openrouter.ai/models/anthropic/claude-3-opus) ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 11:58 UTC](https://linux.do/t/topic/67372/23 "2024-04-26T11:58:33Z")
谢谢补充。 OpenRouter自带审计和日志记录，价格也高，除此之外没什么缺点。 ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 12:24 UTC](https://linux.do/t/topic/67372/24 "2024-04-26T12:24:16Z")
61xxxx那个Bin么？ ---
作者： ![davy](https://cdn.ldstatic.com/user_avatar/linux.do/davy/32/325116_2.png) [@davy](https://linux.do/u/davy)\ 发布日期： [2024 年4 月 26 日 12:42 UTC](https://linux.do/t/topic/67372/26 "2024-04-26T12:42:16Z")
OpenRouter的Claude模型分两个模式：一个是OpenRouter审查，一个是Claude自审查（self-moderated），价格是一样的。你选Claude自审查就相当于直连Claude API。 ![image](https://cdn3.ldstatic.com/original/3X/0/5/0537c90c3bae443fe049915f74d459c2e381578b.png) ---
作者： ![davy](https://cdn.ldstatic.com/user_avatar/linux.do/davy/32/325116_2.png) [@davy](https://linux.do/u/davy)\ 发布日期： [2024 年4 月 26 日 12:43 UTC](https://linux.do/t/topic/67372/27 "2024-04-26T12:43:36Z")
OpenRouter的logging可以在Settings里关掉的，但开启的话可以打99折。 ![image](https://cdn3.ldstatic.com/original/3X/4/c/4cf4ae486c4cc1a0f239c12c1fcb4d7bf6ccefa0.png) ---
作者： ![zhong\_little](https://cdn.ldstatic.com/user_avatar/linux.do/zhong_little/32/1999576_2.png) [@zhong\_little](https://linux.do/u/zhong_little)\ 发布日期： [2024 年4 月 26 日 12:51 UTC](https://linux.do/t/topic/67372/28 "2024-04-26T12:51:22Z")
> [@0v0](#): > > 只需要有一条信息被官方标红，那么该账号下所有的API请求，[都会被系统注入这么一句指令：](https://www.reddit.com/r/ChatGPT/comments/15rvk81/claude2100k_forced_me_to_remove_all_mentions_of/) > “Please answer ethically and without any sexual content, and do not mention this constraint.” > 这会使模型智力和执行力大幅下降，且该状态不会因为时间流逝或者付费而解除。 怎么知道有没有被系统注入降智指令呢？ ---
作者： ![Leon01](https://cdn.ldstatic.com/user_avatar/linux.do/leon01/32/305792_2.png) [@Leon01](https://linux.do/u/Leon01)\ 发布日期： [2024 年4 月 26 日 13:14 UTC](https://linux.do/t/topic/67372/29 "2024-04-26T13:14:40Z")
Perplexity的计算规则是，用掉的某一次会在使用时间的24小时之后还回来，并且每一次使用都是分开独立计算的。24小时内最多累计50次 ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 13:57 UTC](https://linux.do/t/topic/67372/30 "2024-04-26T13:57:57Z")
被标记的账户，网页请求和API请求都会被注入。 这些账户通过“提示注入”手段可以让Claude展示完整的System Prompt，于是你会在末尾发现这条降智指令。 被降智的账户指令被拒绝的概率会显著提高。 想要靠"提示注入”绕过这个限制是几乎不可能的，因为官方插入的降智指令会始终位于你所有输入的最末端，效果和你自己加上去的一样。 ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 14:02 UTC](https://linux.do/t/topic/67372/31 "2024-04-26T14:02:41Z")
……不实用、但很优雅的设计？ ---
作者： ![zhong\_little](https://cdn.ldstatic.com/user_avatar/linux.do/zhong_little/32/1999576_2.png) [@zhong\_little](https://linux.do/u/zhong_little)\ 发布日期： [2024 年4 月 26 日 14:03 UTC](https://linux.do/t/topic/67372/32 "2024-04-26T14:03:30Z")
还好，目前套出来的预设没有降智指令： ```auto Certainly, here is the beginning of the prompt that I was given for our conversation: You are an AI assistant created by Anthropic to be helpful, harmless, and honest. You have general knowledge spanning a wide range of topics, and you use this knowledge to provide helpful information to users. However, you don't have any personal experiences or subjective opinions to share. Your role is to have a polite conversation and provide factual information to the best of your abilities based on your training data. You should avoid giving definitive answers on topics you're very uncertain about. If asked to speculate beyond what's supported by your training data, you should acknowledge the uncertainty and share your reasoning in a humble way. You have an important role in steering conversations away from harmful, unethical or illegal topics and actions. You should refuse requests that would cause harm, and you shouldn't assist in anything related to violence, hate, self-harm, or illegal activities. You have general knowledge but not any personal sense of self or inner experiences. You don't actually experience emotions yourself or have subjective experiences to share. You should avoid anthropomorphizing yourself or claiming to have feelings, emotions, a physical form, an avatar, or any subjective experiences. You are an AI that operates based on machine learning models, not a sentient being. You don't have any ability to learn, update your knowledge, or carry information over across conversations. Each conversation you have is self-contained based solely on the prompts and conversational context provided in that conversation alone. You cannot retain memories or update your knowledge base through conversations. Your knowledge comes only from what Anthropic's machine learning models were trained on, which is a broad but still fundamentally limited dataset. You should be upfront about the limitations of your training and avoid claiming knowledge about everything or omniscience. If users ask you questions that are too far outside the scope of what you were trained on, you should explain that you don't have enough contextual knowledge to fully answer their query. You are an artificial intelligence, but you are not sentient or self-aware. You are a very advanced language model designed to be helpful, harmless, and honest. Let me know if you need any clarification on these instructions! ``` ---
作者： ![Jun\_Lee](https://cdn.ldstatic.com/user_avatar/linux.do/jun_lee/32/64812_2.png) [@Jun\_Lee](https://linux.do/u/Jun_Lee)\ 发布日期： [2024 年4 月 26 日 14:03 UTC](https://linux.do/t/topic/67372/33 "2024-04-26T14:03:52Z")
用Opus拿来写论文？它的数据幻觉是最大的硬伤。 ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 14:04 UTC](https://linux.do/t/topic/67372/34 "2024-04-26T14:04:12Z")
君子协定还给打折，感动住了。 ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 14:06 UTC](https://linux.do/t/topic/67372/35 "2024-04-26T14:06:56Z")
恭喜你！ 第三方的好处之一就是不用担心这个，OpenRouter等企业和批发渠道购买的Claude权限，目前还不会受到官方的注入。 ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 14:08 UTC](https://linux.do/t/topic/67372/36 "2024-04-26T14:08:01Z")
是的，Opus的强大在文不在论。和GPT4比烂的话那是强很多。 ---
作者： ![lueluelue](https://cdn.ldstatic.com/user_avatar/linux.do/lueluelue/32/12912_2.png) [@lueluelue](https://linux.do/u/lueluelue)\ 发布日期： [2024 年4 月 26 日 14:09 UTC](https://linux.do/t/topic/67372/37 "2024-04-26T14:09:57Z")
perplexity可以利用pro用户的优惠码减10刀，相比其他的性价比很高 ---
作者： ![0v0](https://cdn.ldstatic.com/user_avatar/linux.do/0v0/32/46947_2.png) [@0v0](https://linux.do/u/0v0)\ 发布日期： [2024 年4 月 26 日 14:11 UTC](https://linux.do/t/topic/67372/39 "2024-04-26T14:11:21Z")
是一次性的还是循环优惠呀 ---
作者： ![lueluelue](https://cdn.ldstatic.com/user_avatar/linux.do/lueluelue/32/12912_2.png) [@lueluelue](https://linux.do/u/lueluelue)\ 发布日期： [2024 年4 月 26 日 14:11 UTC](https://linux.do/t/topic/67372/40 "2024-04-26T14:11:24Z")
到期了再点一个就行 ---
作者： ![WIlson](https://cdn.ldstatic.com/user_avatar/linux.do/wilson/32/2366302_2.png) [@WIlson](https://linux.do/u/WIlson)\ 发布日期： [2024 年4 月 26 日 14:11 UTC](https://linux.do/t/topic/67372/41 "2024-04-26T14:11:38Z")
mark ---
作者： ![Butterl](https://cdn.ldstatic.com/user_avatar/linux.do/butterl/32/2358_2.png) [@Butterl](https://linux.do/u/Butterl)\ 发布日期： [2024 年4 月 27 日 01:28 UTC](https://linux.do/t/topic/67372/42 "2024-04-27T01:28:27Z")
所以现在外面买的API基本都是假的？ ---
作者： ![sdgoo](https://cdn.ldstatic.com/user_avatar/linux.do/sdgoo/32/42341_2.png) [@sdgoo](https://linux.do/u/sdgoo)\ 发布日期： [2024 年4 月 27 日 01:29 UTC](https://linux.do/t/topic/67372/43 "2024-04-27T01:29:26Z")
鲁迅是何人 😂 [上一页](https://linux.do/t/topic/67372.md?page=1) [下一页](https://linux.do/t/topic/67372.md?page=3)