在DeepSeek Harness的官方代码库中，已经出现了一些新的代码，包括DeepSeek-V5、V5-open、V5-next等等，看样子是新一代大模型了。
这些代码到底是什么意思呢？也不能看到V5就搞个大新闻，普通人不懂代码就直接问问大肥鱼自己，给出的解释是新模型的机制准备好了，但现在的v5是占位。
这套设计的意义就是：DeepSeek可以在服务端改 /models 就把v5开出去，客户端不用发版。deepseek-v5-open走"无版本门槛"这条，DeepSeek-v5-next 走"必须升级客户端"这条，正好把两条下发路径都覆盖了。
直白说一点就是DeepSeek官方已经把DeepSeek V5系列大模型的发布做了一些提前的准备，现在就用占位符占了位置了，未来发布之后服务端只要改/models就能发布，而客户端之后都不用发新版就能支持。
目前的代码并不能说明DeepSeek很快就要发V5版大模型，但可以证明越来越近了，已经开始为发布做铺垫了。
DeepSeek的大模型命名也一直比较迷，9月17日发布的V4.1 Flash已经是完全换新架构了，跟之前的V4没有可比性了，结果命名还沿用了4.0系列的，1017发布的版本提升也很大，结果也没变成V4.2。
考虑到之前的V4 Pro正式版不太美好的回忆，DeepSeek直接把新一代旗舰大模型改成DeepSeek V5 Pro也是不错的办法，Flash版刚发布没多久，不过V5 Pro的发布也不会很快，本月应该没戏，要到11月份了。
如果是11月中旬，那DeepSeek V5 Pro、V5 Flash倒是可以正好一起换新，命名也不会那么混乱了。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
In the official code library of DeepSeek Harness, a number of new codes have emerged, including DeepSeek-V5, V5-open, V5-next, etc., which appears to be a new generation of LLM.
What do these codes mean? Neither can V5 make a big story, and ordinary people ask the big fish themselves if they don't know the code.
The meaning of this design is that DeepSeek can change at the service level/models and drive V5 out without a copy. Deepseek-v5-open takes the "no version of the threshold" and DeepSeek-v5-next takes the "must upgrade the client" to cover both of the downways.
To put it bluntly, DeepSeek official has prepared some advance publish for the DeepSeek V5 series LLM, and now has a placeholder. The future publish service after publish can be publish if it changes/models, and the client does not need a new version to support it.
The current code does not suggest that DeepSeek will soon issue V5 version LLM, but it can be shown that it is getting closer and has begun to prepare for publish.
The name of LLM in DeepSeek has also been fascinating, with V4.1 Flash of publish on 17 September having completely changed its structure, no comparison with the previous V4, the results of which follow the 4.0 series, and the version of 1017publish having been significantly upgraded, which did not turn into V4.2.
In view of the unwelcome memories of the previous V4 Pro edition, DeepSeek’s change from a new generation of flagship LLM to DeepSeek V5 Pro would also be a good idea. The Flash version was just publish, but V5 Pro’s publish would not be very soon, and should not be in play this month, until November.
If it were mid-November, DeepSeek V5 Pro, V5 Flash could have changed it together, and the name wouldn't be so confusing.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.