<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs" -->

---
title: moonshotai/Kimi-K3 · Upcoming release · Hugging Face
description: Moonshot AI is preparing to release Kimi K3, described as the world&#x27;s first open 3-trillion-parameter-class (3T-class) frontier model. It features a new...
canonical: https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: moonshotai/Kimi-K3 · Upcoming release · Hugging Face | daily.dev
og:description: Moonshot AI is preparing to release Kimi K3, described as the world&#x27;s first open 3-trillion-parameter-class (3T-class) frontier model. It features a new...
og:url: https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs
og:image: https://api.daily.dev/og/posts/l6A62mUXs.png
og:image:alt: moonshotai/Kimi-K3 · Upcoming release · Hugging Face
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# moonshotai/Kimi-K3 · Upcoming release · Hugging Face

**[Hacker News](https://daily.dev/sources/hn)** · 1 min read · 2 upvotes · 5 comments

## Summary

Moonshot AI is preparing to release Kimi K3, described as the world's first open 3-trillion-parameter-class (3T-class) frontier model. It features a new architecture based on Kimi Delta Attention and Attention Residuals, native agentic capabilities including tool calling, browsing, and multi-step planning, and an extended context window for repository-scale code understanding. The model weights will be released openly on Hugging Face, with 1,779 users already waiting for the drop as of the expected release date of July 27, 2026.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://huggingface.co/moonshotai/Kimi-K3>

## Community take

How the wider developer community reacted, aggregated from 1 discussion and 145 comments across hackernews (as of 2026-07-27).

**TL;DR:** The community is broadly excited about Kimi K3's open-weights release as a frontier-level model, but discussion is dominated by practical concerns: the enormous hardware requirements (~1.5TB VRAM or RAM), inference speed on consumer/prosumer hardware, API pricing economics, and whether third-party providers can serve it profitably.

**Sentiment:** 45% positive · 35% mixed · 20% skeptical

**The case for**

- Open-weights release of a frontier-level model is seen as historically significant, putting top-tier capability in the hands of anyone with sufficient hardware.
- MoE architecture means only ~32B active parameters need to be in VRAM at once, making expert-offloading approaches feasible for smaller setups.
- MXFP4 native format means it fits on 8x AMD MI355X nodes, which could drive token prices down.
- Competition among third-party providers is already pushing prices down rapidly (e.g., GLM 5.2 dropped ~45% since release).
- Distillation into smaller consumer-friendly models is seen as a realistic downstream opportunity.

**The pushback**

- ~1.5TB VRAM requirement puts it far beyond consumer or even most enterprise hardware.
- CPU/RAM-only inference would likely yield sub-1 tok/s speeds, making it practically unusable even for async tasks given K3's heavy reasoning token usage.
- MXFP4 format is not yet fully integrated into the inference ecosystem (llama.cpp, etc.), limiting who can even run it.
- Concerns that providers may silently serve degraded quantizations without disclosure.
- The intersection of 'needs absolute data privacy, needs SOTA model, cannot afford GPUs' is argued to be nearly empty, limiting the self-hosting use case.

**By community**

- hackernews (mixed): Enthusiastic about the open-weights milestone but heavily focused on practical hardware/cost barriers, inference speed limitations, and API pricing economics rather than pure capability celebration.

**Hottest debate:** Whether running Kimi K3 on CPU+RAM servers is at all practical given its heavy reasoning token usage, with estimates ranging from 'usable for async overnight tasks' to 'so slow it's completely unusable.'

**Open questions**

- What will median $/MTok pricing settle at for a 3T model across third-party providers, and does it reveal whether labs subsidize API pricing?
- Will quantization tools like unsloth produce viable Q8 or Q4 variants, and how much capability is lost?
- Can expert-offloading or hybrid GPU+RAM approaches make single-node inference practical for small/medium enterprises?
- Will Cursor or other coding tools fine-tune K3 the way they did with Kimi K2.x (Composer series)?
- How does K3 perform on censorship/political bias tests compared to other Chinese open-weights models?

**Highlights**

> This will be interesting for a few reasons. First, depending on where the median pricing settles w/ 3rd party providers will tell us what it costs to serve a 3T model. Since it's going to be mxfp4 native, it'll take ~1.5TB of VRAM to host this, which is juuust at the limit of 8xb200s (but realistically you'll need 16x for context / throughput optimisation). Won't be cheap to host, but at least we should get some range of $/MTok for a 3T model. Then we'll be able to guesstimate if "labs are subsidising tokens on API pricing". Also interesting to see what effort it will take to fine-tune this beast. The latest AISI benchmarks on cybersec place it above glm5.2, but still way way behind SotA closed models. Some fine-tuning might be needed here. Also, interesting to see if Cursor does another training round on it, to directly compare it w/ kimi2.6/2.7 fine-tunes (composer series) and grok4.5. Also also, interesting to see if someone takes on distilling (proper distillation, w/ training the entire distribution) from this into smaller models. (dsv4-kimi should be really good, since dsv4 is very cheap to serve)
> — [NitpickLawyer on hackernews · 10 comments](https://news.ycombinator.com/item?id=49065868)

> It will be very interesting to see what kind of 'slow' performance people get from running it on a no GPU, but tons of RAM server (like a dual or quad socket xeon with 1.5 to 3TB of RAM). For the purpose of giving it longer duration tasks to generate a piece of something and come back and check on what it has done in 4 or 6 hours. Even if the output is like 5-6 tok/s, that might be usable for some purposes. Huge price difference in what you can do with buying a used 4U rackmount server and putting 3TB of RAM in it (64GB DIMMs x quantity 32 in a quad socket xeon, you can see some benchmark prices on eBay for sets of 16 or 32 matched 64GB ECC DIMMs) for <$30,000, vs the cost of trying to run it on real GPU hardware. Now obviously, as of the time I write this, the full precision hasn't been released nor has anyone like unsloth run it through quantization yet to produce a "Q8" or "Q8-XL" variant of it. But I think it's going to need more than 1536GB of RAM, with a usable and large amount of context, more like 2TB and preferably 2.5 to 3TB. I also predict that people who try to run it in Q4 and Q6 will get the worst of both worlds, less precision/lost knowledge but also not reliable output that comes out too slow. In my personal opinion if I'm going to deal with something that is smart but slow and running on limited budget hardware, I need it to be Q8.
> — [walrus01 on hackernews · 4 comments](https://news.ycombinator.com/item?id=49066583)

> Lets say an easy response takes 32k tokens in total, and to be generous, let's say it does 1 tok/s. This is already ~9 hours, and 32k reasoning tokens isn't even that much and as mentioned, K3 probably does the longest/most reasoning/thinking out of the available open weights models today, much like GLM. Just lowering that performance to 0.5 tok/s, would lead to ~18 hours for a simple prompt to receive an answer. And then that's just for single prompts, what about agent harnesses, where before every tool call the model could reason a bunch? I agree with you that real world results would be interesting, but I wouldn't hold my breath nor expect it to realistically be able to be useful. Still, people should try it, for science if nothing else :)
> — [embedding-shape on hackernews](https://news.ycombinator.com/item?id=49066824)

> I feel like most hardware to run LLMs on is shaped wrong for individuals. It's either having a model struggling along with like 5-10 tokens per second on unified memory, or data center cards with hundreds of GB of VRAM consuming more than a kW of power. It doesn't seem like there's prosumer GPUs with like 180W-250W TDP and 128 GB or 256 GB of VRAM (one can dream). Then bifurcation and even just two of those cards would be kinda useful (albeit NVLink or equivalent would need to be commonplace). Obviously nobody is running Kimi K3 locally without an insanely beefy homelab and lots of money to burn, but running GLM 5.2 would be cool at like ~100 tokens per second for a single session and maybe ~60 tokens per second with N subagents. How unfortunate.
> — [KronisLV on hackernews · 5 comments](https://news.ycombinator.com/item?id=49066552)

> This is historic. For the first time, an open-weights LLM is right at the top. We won't be able to run this ourselves, but many providers can.
> — [davidkunz on hackernews · 1 comments](https://news.ycombinator.com/item?id=49066201)

**Source threads**

- [hackernews](https://news.ycombinator.com/item?id=49065752) · 117 points · 145 comments

## Community discussion

Top comments from developers on daily.dev.

**@hardik15** · 1 upvotes

> This open model is the reason, this properitary models needs to push hard to release their new models, specially Anthropic and OpenAI.
>
>
> By any chance this models are becoming better than theirs and once people done hardware setup, their downward sales trend can starts.
>
> That's why they are pushing hard to make this happen

**@doncho** · 1 upvotes

> I'm looking forward to testing this written a real project.
>
> So far, I'm very lazy and prefer using the out of the box Claude Code integrations. Couldn't find such easy way for setting up other models.

---

Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#kimi-k3](https://daily.dev/tags/kimi-k3)

[View this post on daily.dev](https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"moonshotai/Kimi-K3 · Upcoming release · Hugging Face","url":"https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs"},"datePublished":"2026-07-27T07:54:57.164Z","dateModified":"2026-07-27T15:45:30.043Z","description":"Moonshot AI is preparing to release Kimi K3, described as the world's first open 3-trillion-parameter-class (3T-class) frontier model. It features a new...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/38295d7cd09934ca294c330e110c3ac5?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/38295d7cd09934ca294c330e110c3ac5?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hacker News","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":5,"discussionUrl":"https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":5}],"keywords":"llm,ai-agents,kimi-k3","timeRequired":"PT1M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"moonshotai/Kimi-K3 · Upcoming release · Hugging Face"}]}
{"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/moonshotai-kimi-k3-upcoming-release-hugging-face-l6a62muxs","comment":[{"@type":"Comment","text":"This open model is the reason, this properitary models needs to push hard to release their new models, specially Anthropic and OpenAI.\nBy any chance this models are becoming better than theirs and once people done hardware setup, their downward sales trend can starts.\nThat’s why they are pushing hard to make this happen","datePublished":"2026-07-27T08:06:10.641Z","url":"https://daily.dev/posts/l6A62mUXs#c-LWu1mT6BV","author":{"@type":"Person","name":"Hardik","url":"https://daily.dev/hardik15","image":"https://media.daily.dev/image/upload/s--O0TOmw4y--/f_auto/v1715772965/public/noProfile"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}},{"@type":"Comment","text":"I’m looking forward to testing this written a real project.\nSo far, I’m very lazy and prefer using the out of the box Claude Code integrations. Couldn’t find such easy way for setting up other models.","datePublished":"2026-07-27T09:02:04.811Z","url":"https://daily.dev/posts/l6A62mUXs#c-KPTwmQ2ls","author":{"@type":"Person","name":"Doncho Angelov","url":"https://daily.dev/doncho","image":"https://media.daily.dev/image/upload/s--j1MSm0hy--/f_auto/v1716369534/avatars/avatar_EB0j0Umzh5v3zGWUSL5ma"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}}]}
```

