MoonShot AI COOKED: Kimi 2.6 + Kimi Code ARE NUTS

Income stream surfers| 00:12:03|May 10, 2026
Chapters22
The speaker argues that people are misusing Kimmy and suggests the correct approach is to use Kimmy inside Kimmy itself.

MoonShot AI’s Kimmy 2.6 shines as an affordable, fast CLI-driven alternative to Claude Code, with eye-catching hands-on testing and a clear plan for future model comparisons.

Summary

Income Stream Surfers’ Mish walks through Kimmy 2.6 and the Kimmy CLI, arguing it’s the most practical way to leverage Kimmy’s capabilities today. He highlights a $19/month Moderato plan offering around 42-43 tokens per second and demonstrates the system inside Kimmy Code. Mish compares Kimmy 2.6’s pricing to Gemini 3 Flash, Anti-gravity benchmarks, and GLM 5.1, noting Kimmy’s cost-per-token dynamics and potential advantages for cost-conscious developers. He also touches on Miniax CLI options and signals upcoming tests against Mini Max, Deep Seek, and other models like Opus 4.7, Gemini, and Quen. The video blends live UI exploration (Italian vs English prompts, a multi-block homepage design) with blunt opinions about model quality and speed, all while promising ongoing model rankings. He even peppers in a plug for Harbor SEO AI, framing it as a companion tool for content ranking and site health. The tone is brisk, experimental, and future-looking, with Mish planning to publish further model comparisons in quick succession. Expect a practical, test-driven take on how Kimmy 2.6 stacks up against current giants and how CLI-first usage can unlock cost-effective AI engineering today.

Key Takeaways

  • Kimmy 2.6 pricing shows input tokens at 0.75 million and output tokens at $350 million, making it a competitive option compared to Gemini 3 Flash and other cheap models.
  • The $19/month Moderato subscription is highlighted as a cost-effective way to access Kimmy Code with around 42-43 tokens per second throughput.
  • Mish demonstrates Kimmy 2.6 inside Kimmy Code, emphasizing the speed and a near-final product quality in a real-world, premium-drag-and-drop UI scenario.
  • He notes a design/engineering gap relative to Opus 4.7 (design and detail) but still rates Kimmy 2.6 an 8/10 on his personal scale, reserving a 10/10 for Opus 4.7 with Claude Code.
  • The video signals ongoing, rapid testing plans across multiple models (GLM 5.1, Mini Max, Deep Seek, Quen variants) to build a running leaderboard.
  • Harbor SEO AI is promoted as a monetizable tool for ranking content with a focus on CMS integration and AI-assisted rewriting and page generation.

Who Is This For?

Developers and AI enthusiasts who want practical, hands-on testing of Kimmy 2.6 and a realistic plan for comparing cheap AI models against major players like Gemini, Claude Code, and Opus.

Notable Quotes

"Look at this guys. In my opinion, everyone is using Kimmy incorrectly. The actual way to use Kimmy is inside Kimmy CLI, which is the very, very cheap version of Claude Code that you can use today released by Kim."
Mish sets up the central claim that Kimmy’s CLI approach is the optimal way to deploy Kimmy.
"Kimmy 2.6 is 0.75 million input tokens, $350 million output tokens. Not actually sure how this compares to, you know, everyone, but for example, compared to Sonnet or Opus, it's extremely cheap."
Key pricing snapshot used for early model comparison.
"This is my favorite Chinese model at the moment. Moonshot is cooking something over there."
Subjective assessment that positions Kimmy 2.6 as a standout option.
"8 out of 10 for sure. And Opus 4.7 with Claude Code is a 10 out of 10."
Mish assigns a personal ranking that sets expectations for future tests.
"If you want to lock in this pricing, go and get a free trial right now and lock in this pricing and I will never raise the price on you."
Promotional note tied to Harbor/creator sponsorship and pricing.

Questions This Video Answers

  • How does Kimmy CLI compare to Claude Code in real-world tasks?
  • What is the pricing breakdown of Kimmy 2.6 for tokens and throughput?
  • Can Kimmy 2.6 realistically replace more expensive models like Opus 4.7 in a production workflow?
  • What models will Mish test next after Kimmy 2.6 and GLM 5.1?
  • Is Harbor SEO AI a viable alternative for content generation and site health optimization?
MoonShot AIKimmy 2.6Kimmy CLIClaude CodeGemini 3 FlashGLM 5.1MiniaxOpus 4.7Deep SeekHarbor SEO AI
Full Transcript
Look at this guys. In my opinion, everyone is using Kimmy incorrectly. The actual way to use Kimmy is inside Kimmy CLI, which is the very, very cheap version of Claude Code that you can use today released by Kim. And that's exactly what we're going to be talking about in today's video. Let's jump into things. So, a lot of people talk crap about Kimmy because it's slow, but if you actually look at this system here inside Kimmy Code, and this is using my subscription, right? So, this is using a $19 a month subscription which you can get today from their website and it's extremely fast, right? 42 43 tokens per second which, you know, it's not quite up there with the extremely fast models, but it is still very very fast. So, this is my subscription here. You can see I am on the moderato moderato subscription and it is $19 a month and so far I've used 1.56% of this uh subscription. So this is from the start of this project that I'm going to show you everything of. This is just the school community prompt that creates basically a servicebased website. But I just want to show you the difference between this and you know kind of the other cheap models on the market. Now, let's just jump on over to uh Open Routter just to look at the pricing of Kimmy K. So, Kimmy 2.6 is 0.75 million input tokens, $350 million output tokens. Um, not actually sure how this compares to, you know, everyone, but for example, compared to Sonnet or Opus, it's extremely cheap, right? Which is always good. It's kind of I think it's like Gemini 3 flash pricing. Let's just have a look. Yeah, it's about Gemini 3. It's slightly more expensive than Gemini 3 Flash preview, which in my opinion is an extremely good model. So, this is kind of what I would be comparing it to. If you do this build inside Anti-gravity, which is free, with Gemini 3 Flash, it does do a pretty damn good job, I have to say. Now, let's compare this to GLM 5.1, which is what uh kind of a lot of people would compare it to. So, it's it's actually going to work out potentially slightly cheaper because normally you use more input tokens than output tokens. So, again, very very interesting here. I probably will do some more tests on all of the latest models. This is my test of Kimmy 2.6, but I will also do a test of GLM 5.1 and a few others as well. Let's see what Miniax is saying. I actually I have to say I really I'm not a fan of Miniax to be honest. Um that's very very cheap though. Miniax is 030 per million input tokens and $120 per million dollar alpha tokens. Now, do Miniax have a CLI? They do have a CLI. So, I haven't actually looked into this for token plan users. No coding is required to unlock minimax motion video generation. Okay, so you can do video generation, speech synthesis, music creation, coding, and more. You can call these capabilities directly in assistance like open claw and claw po. That's pretty crazy. So, yeah, this video that I'm doing right now is about Kimmy 2.6, but I do look at all models and I do kind of test all models. So, I'll be curious to see um how good the Miniax CLI is as well. This is probably the next test that I will do. It's pretty interesting. You can do uh speech synthesis, image generation, video generation, music generation, web search, image understanding. So, that's definitely a plus. Um, it'd be pretty cool to see if you could like generate an entire website with like a hero video, but that's definitely for a video that's coming in the future. Okay, just before we see the result of the video and just before we continue with the video, huge shout out from our sponsor, me. This is harbor seo.ai, which you can find a link in the description and in the pinned comment. This is basically the tool that I've been putting the most effort into over the last few years. It is an SEO tool, a GEO tool. It will help you rank in LLMs and inside Google. And if I just go and show you what it actually looks like. If I go to dashboard here and I go to the top and go to billing and plan, you can see that we currently have in total 377 published pages, 172,000 impressions from those pages, which is kind of absurd. And then almost 1.7K clicks as well. Now, you can go and get founder pricing. I will like Sorry. You can go and go get this pricing here. I am putting the pricing up soon when I release my website builder because I need to basically because the API costs are so high. So if you want to lock in this pricing, go and get a free trial right now and lock in this pricing and I will never raise the price on you. Okay? You can go and get free trial right now. Go and try it out. We just released Copilot which basically allows you to generate all of this content from like a messy CSV. So this is just a copied and pasted CSV and it basically just made all this content for me right from this CSV. Uh we've got the dashboard, the main dashboard where you can run site health scans. You can fix it with AI automatically which will automatically you know do all the metadata for your website. We have the discover tab where you basically find keywords and then you can write the keywords or build the pages. And then we have the main thing which is the writer which has this content generator, landing page generator, rewriter which rewrites your current articles. And we have this new one which is called the free form which basically allows you to use your prompts to write the content your way that still uses our scraping techniques. So this is Harbor, guys. Go and check it out in the description. You will not find a better tool for the price. If you looked for a thousand years, I promise you this is the best tool on the market for the price. Thanks for the attention. Let's jump back into the video. Okay, so my plan is to kind of make like a running tally of all of these models, right? So, we'll start with uh Kimmy 2.6 today. This is by Moonshot AI. And I will just quickly say this is probably my favorite Chinese model at the moment. I have I I don't know what they're doing over there at Kimmy, but they're definitely cooking something, right? And then we'll do a test of GLM 5.1 probably tomorrow. This is by Z.AI. AI. They did actually sponsor the channel, so I owe them a video, even though I've already made loads of videos about them. Um, so yeah. Uh, and then let's do Mini Max. I think it's 2.7, which is I think the company's called Miniax, but don't quote me on that. Uh, this is my least favorite at the moment, but like I'm not going to let that bias anything. If this performs better than this, then I'll move it up in the rankings. Right. If you want to see any other models, just leave a comment and say, "Hey, Mish, can you also cover this model?" And I will I will look into it. Um, I'll probably compare it to like a baseline of like I would say Gemini 3 Lash probably. Oh yeah, I'll do another test with Deep Seek as well. Um, I actually really like Deep Seek V4. I think it's kind of groundbreaking and I wouldn't be surprised if it dragged the rest of these models up with it. Not Gemini 3 Flash, obviously. There are some other interesting models. There's Gemma 27 or 31B, which I'll probably do some tests on as well. There's Quen uh I think it's 27 uh so 3.6 27B. I can't remember if it's 27 to 31. Uh which I'll also do this test on. Um let's see what else there is. I think most people are waiting for the next generation of these models. I think people are quite excited about the next generation of these models. As I kind of alluded to already, in my opinion, Kimmy 2.6 by Moonshot is kind of the best of the best. And they're definitely cooking something over there. And I I would say the next Kimmy model is going to start to actually, you know, not compete with Opus. I won't say compete with Opus. It's a bit of a meme at this point, but like it's getting there. Okay, guys. So, kind of this is what I'm talking about. Not only is the quality of this extremely good, like the um the design, like I really really like the design here. It looks very similar to another design that I got from this exact prompt actually. But you know, we don't just look at design here on Income Stream Surfers. We look a little bit beyond the just the design into the technical build. So if I click Italian here, you can see it changes to Italian. There's a slight error there. Just so we all know, it's not actually finished yet. This is more like a preview, but like I can already tell that the um the quality is amazing here, right? Let's just see if I click en it. Yeah, it changes properly. Let's just see here. So, we've got um let's see if we click here. Okay. Rolls-Royce uh luxury car hire or Campa, whatever. Doesn't matter the pronunciation. Um Ariano Pino, beautiful. Look at that. Really, really good quality. Let's see what happens if I press book now. Amazing technical build. It's pretty much perfect. Okay, so there's a client side error there. That's likely due to the fact that it hasn't actually finished. I don't normally do this until it's actually finished. Um, but like it's very very close to being finished. Like I'm not I'm not really missing out on much by just waiting an extra like 15 20 minutes for it to fully finish. One of the main things is if I click all of these in the footer, does it work? It seems like it is working, right? You can switch to Italian. You can press book now. Contachi. You can press campa. I mean this is literally a perfect technical build from Kimmy K. Right. This is, you know, really, really impressive stuff. Right. So, as a baseline for Kimmy, I think what we're going to do is I can't give it a 10 out of 10 because if you start with 10 out of 10, you've got nowhere to go. I would like the design to be a bit better. If you did this with Opus 4.7, the design would be better. It would also be a little bit more detailed. I would say it's lacking a little bit of detail. So, we've got one block, two blocks, three, four, five, six, seven. The the prompt actually says 5 to seven horizontal blocks. So, I mean, honestly, I would give this a 10 out of 10, but I don't want to start on a 10 out of 10. And what I mean by 10 out of 10 is like 10 out of 10 on my own scale, right? If you're wondering what's causing those issues, it's uh basically it's still coding in the background there. So, uh that's actually what's happening. Um I will let this end, right? I will let this finish just in case it decides to, I don't know, do something crazy at the very end. But just generally speaking, this is an incredible build. Very, very happy with the result. And I'll give this an 8 out of 10 just so that we have a good base level, right? Um, and we'll say that Opus 4.7 with Claude Code is a 10 out of 10. Right? So, this is my scale. You can say whatever you want about it. But yeah, basically my next few videos, guys, are going to be about this topic. So, be on the lookout for upcoming videos if you enjoyed this one. That only took about 15 to 20 minutes as well. So, the speed is pretty good. people that are saying the speed is bad, they're probably not using Kimmy's own API. You have to actually use their API before you make a comment in my opinion. But yeah, I mean overall very very impressive build. I will uh I'll I'll make sure that nothing else big happens before I end this video. But with all that being said, guys, if there is nothing else after this, then it's finished. And I'm very very impressed. 8 out of 10 for sure. Thank you so much for watching. If you are watching all the way to the end of the video, you're an absolute legend. And I'll see you very very soon with some more content, probably with the uh GLM 5.1 video tomorrow. Uh which you can get I'll I'll leave a link at the end of this video. So if you do want to watch the rest of the series, go and check out that video. Thanks for watching, guys. Go and check out Harbor to support me. And as usual, peace out.

Get daily recaps from
Income stream surfers

AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.