DeepSeek V4 Pro can be integrated into Claude Code using Claude Code Router (CCR) with a $20 subscription, offering a 1M context window and 1.6T MoE parameters at $0.435/M input and $0.87/M output costs, though it scored 7/10 in testing due to higher-than-expected costs and some technical limitations compared to premium models like Opus 4.7.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
DeepSeek V4 Pro Inside Claude Code = GOD MODE (FREE?)Added:
Okay guys, so let's continue our series that we're doing right now trying to find the best and most cost effective way to code in 2026. Okay, so today is the turn of DeepSeek. I'm going to be using it inside Claude Code. You can use something called Claude Code Router.
There's Claude Code router. You'll have to work this out for yourselves, guys, but you can get Claude Code to set this up for you, just so you know. Uh it's kind of meta, but the way this works is you can write CC R. So claude code write and then uh code so CCR code and then set the model to DeepSeek and then you're actually going to be using DeepSsee inside quad code which you can do with just the $20 subscription.
Just to show you guys if I go to logs you'll see that the last kind of conversation that I had was with Claude Code uh and Deepseek. So Deep Seek V4 you can see right here. Okay. So yeah, this is going to be Deepseek V4 Pro inside Claude Code. So let's get the test started. Okay, so this is just the normal prompt I do for all of these tests. We're just going to put this here and we'll try and get an overall cost of this as well. Um but it might be kind of hard to get a cost, but I should be able to work it out. Maybe I can do um model deepseek here and then it'll give me an overall kind of uh idea of the cost. Okay, so he decided to enter plan mode himself or itself, whatever. I just say him. I don't know why. Um, so it's currently reading everything. It's entering plan mode. I did not tell it to do this. Uh, normally I would use the superpowers plugin, but it's decided to enter plan mode itself. So, we'll see how this affects the overall build. Okay, so this is the plan that it came out with. I'm not going to change anything. I'm just going to say go for it, and we'll see how this does.
Really, really interesting that it actually decided to do a plan. I've got a tracker on how much this is going to cost. And yeah, I am playing Runescape in the background. You can ignore that.
But yeah, we'll see what happens here.
And uh yeah. Okay, guys. So, just for some context, um the video I did yesterday, I gave Kimmy inside uh Kimmy CLI an 8 out of 10. I will probably test this inside Claude Code as well. My basic idea is to test all things in all places. So, I'll also do an open router video of Deep See V4, for example. I just want to give people kind of an an understanding of what they should be using. But if if I were you and you didn't have the budget to play to pay for the $200 uh Claude Max subscription, this is like an alternative, right? An a alter. Okay, come on. I must be able to spell alternative, right? Uh this is like a a video series trying to show alternatives to the $200 max subscription. So, right now, I would say that your best idea would be Kimmy 2.6 with Kimmy CLA if you're not paying the $200 or $100 Claude Max subscription.
But we'll see how Deep Seat V4 and Claude Code compares to Kimmy CLI and Kimmy 2.6. So, that's kind of where I'm going with these videos just so everyone knows. Okay guys, so this has cost more than $7 just so everyone knows. So, it is actually fairly expensive to use the Deep Seek Pro like this. I guess that's Claude code using these credits. Okay guys, just before we continue and we see the final result here, quick shout out from our sponsor, me. This is harbor SEO.ai, which is a writer and blog content generator, which I created. I am shipping updates constantly on this thing literally every single day. I read every single message that people write in the bottom right corner here. And if there's an issue or something like that, I fix it immediately. For example, there was an issue with the WordPress connection. People were just struggling to basically connect WordPress so that they couldn't so they could post content automatically. I just fixed this. So now there is a plug-in which you can easily download. And the person that complained about this is now very very happy because it's working. If I just go to billing and plan right here, I'll show you the results that we've got so far.
387 pages published, 180,000 impressions and almost 1.7,000 clicks. I'm extremely happy with how things are going. If you do want founder pricing, guys, feel free to sign up. Use the free trial, which is a 3-day free trial, and then message me again, bottom right corner, I see every message, and just say, "Look, can I have the founder pricing, please?" And I'll make sure that you get that. Like I said, there is a 3-day free trial, so there's really no excuse to not go and give Harbor a try. Okay, guys. So, we are done. Uh, let's have a little look at the result here. So, first thing I would say, the design is not that good.
Another thing is the SVGs are definitely a bit off, but one good thing about this model is it has 1 million context and it was actually quite expensive to build as well. To be honest with you, this cost about 12 bucks. I don't think deepseek is necessarily the best but what I do think is that this is the beginning of the free not free open source or open weight Chinese models starting to have 1 million context and starting to be uh a little bit better and also more efficient. Right? I think that this is a step forward for these LLMs, but I don't think that it's going to come from Deep Seek V4 Pro. I think it's going to come from whoever builds with Deepseek V4 Pro as their base. Right. I have to say as a technical build, it's pretty good. Um, it it built for a long time as well. I don't know what this is. For some reason, this is not working.
Uh, there's like a problem with when you swap to Italian where it doesn't properly swap, but that's okay. Um, I'm going to give this I think, let's see, I think I'll give this a 6 out of 10. Um, it's actually quite disappointing. I expect this to be better to be honest with you. But if you can run this on your machine, I would probably say that this is one of the better models, but honestly, I'm looking forward to what comes next, right? So, that's specifically why I'm looking here with Deepseek. I have a theory, and I don't know if this is right or not right or whatever. I think these three models are actually based on DeepSeek, like the original V3 Deepseek that, you know, kind of blew up the internet when when that happened. Um, because then these three models kind of came out of nowhere after that was released. So, if Deep Seek leaps forwards with efficiency, these three models should also leap forward with efficiency. I'll be particularly looking at Kimmy as probably the main one, although people normally talk about GLM as well. And then Miniax for me, I mean, it's very very token efficient, but it's definitely my least favorite uh Chinese model. There are a few other models that I do like to use and that we will test.
And also, I will do a test of the nonpro version. So, the light version of Deep Seek. So, let me just add the word pro here because this is the Pro version uh that we just tested and I do want to do a test on the nonpro version um because it was actually pretty good when I tested it last time. So, however, with that all being said, guys, I think I'll leave the video there. Deepseek 6 out of 10. Not bad, but I think it could have done better. I think we'll try GLM 5.1 next. Um, and then I do want to do Deep Seek V4 Light as well because it does seem like a fairly good model. But yeah, I mean, I don't know the maybe I'm being a little bit harsh on this model. Maybe this design is better than um actually I might I might change it to to 7 out of 10 to be honest with you. The only problem is some of the technical stuff isn't there, which is a bit of a shame to be honest with you. But overall, it is pretty good. I think I'm going to change it. Actually, guys, I think I'm going to give it a 7 out of 10.
Okay, so we have our kind of um leaderboard right uh now. Like I said in yesterday's video, Opus 4.7 is a 10 out of 10 for me, right? So, we'll do a test with Codeex as well, but yeah, for now, this is what we're looking at, guys.
I'll leave the video there. This is a decent model. Uh, it is quite expensive though, so just be wary of that. Uh, thank you so much for watching. If you are watching all the way to the end of the video, you're absolutely legend. Go and check out harbor seo.ai and I'll see you very, very soon with some more content. Peace out.
Related Videos
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30











