First screen, captured 9 days ago. Open the video.

Kimi K3.1 Pro model weights are out

What HN thinks

Open weights for everyone, provided everyone owns a B300 node and a spare power plant.

Too big to run locally
Most commenters note the 2.8T model needs datacenter hardware, though some describe SSD expert-streaming or offloading tricks that make it crawl on big-RAM machines.
Benchmaxxing suspicions
Some argue K3.1 doesn't beat DeepSeek V4.1 Flash by much and cite an audit suggesting training on benchmark test sets, while daily users insist it feels much stronger than Flash.
Open weights still matter
Even if unrunnable at home, the weights enable provider competition, on-prem enterprise use and cost savings over Claude, with the license only biting large model-as-a-service resellers.

Consensus: Commenters agree this is a strong open model a bit behind the frontier that hobbyists can't realistically run; its benchmark honesty is disputed.

Recommended in the comments

OpenRouter ↗ (service): Lets you compare prices and route to cheap or zero-data-retention providers for open models.
Together AI (service): Self-hosts open models with data retention off and reportedly higher throughput than Moonshot's endpoint.
oMLX (tool): Supports MTP, n-gram table offload and expert offloading to fit big MoE models on Macs.
Strata (tool): Users report running large quantized MoE models well on 16–24GB VRAM plus lots of RAM.
Qwen3.8-27B (library): Recommended as a capable model that fits on consumer GPUs when quantized.

Summarised by Claude from the comment thread 8 days ago.

Points: 607
Comments: 174
Best rank: #6
Posted: 2026-11-04 by abrarthought
Category: AI & ML

kimi moonshot-ai open-weights llm mixture-of-experts hugging-face multimodal

All captures:

Full page

Captured 9 days ago. Click for the whole screenshot.

Looks similar

Screenshot of github.com: Peer-to-peer WebRTC connections with built-in signaling and discovery

Peer-to-peer WebRTC connections with built-in signaling and discovery (github.com)

56 points by LuKks 5 days ago | 17 comments

GitHub repository for like-webrtc, a JavaScript library for peer-to-peer WebRTC connections with built-in signaling, cryptographic identities and peer discovery.

webrtc peer-to-peer javascript npm

Screenshot of github.com: 6 Hours of YouTube and 85 handtyped pages of notes on LLM Inference – Start here

6 Hours of YouTube and 85 handtyped pages of notes on LLM Inference – Start here (github.com)

56 points by samirswr 4 days ago | 9 comments

GitHub repo for a small LLM inference engine built from scratch in C++ and CUDA, with a YouTube video series and 80+ pages of notes.

llm-inference cuda cpp kv-cache

Screenshot of github.com: Huawei releases MACaQ, its internal quantizer for local inference

Huawei releases MACaQ, its internal quantizer for local inference (github.com)

121 points by thefox96 2 days ago | 30 comments

GitHub repository for Huawei CSL's MACaQ, a model-agnostic quantization tool for text, image, video and audio models targeting llama.cpp, vLLM, Diffusers and PyTorch.

quantization gguf llama-cpp vllm

Screenshot of ycombinator.com: Mbodi AI (YC P25) Is Hiring Research and Robotic Engineers

Mbodi AI (YC P25) Is Hiring Research and Robotic Engineers (ycombinator.com)

0 points by chitianhao 7 days ago | 0 comments

Y Combinator jobs page for Mbodi AI (YC P25), an embodied AI robotics startup hiring a Founding Research Engineer and Senior Robotics Engineer in New York.

robotics embodied-ai y-combinator hiring

Screenshot of lists.trustedfirmware.org: Future Maintenance of Mbed TLS

Future Maintenance of Mbed TLS (lists.trustedfirmware.org)

61 points by jonesjohnson 10 days ago | 17 comments

Mailing list announcement that current Mbed TLS maintainers will stop maintaining the TLS and X.509 libraries from April 2027 unless new maintainers step in, with user replies.

mbed-tls tls x509 tf-psa-crypto

Screenshot of windymapy.com: WindyMaps – Free offline maps, 1h/1km weather and route trip planner

WindyMaps – Free offline maps, 1h/1km weather and route trip planner (windymapy.com)

8 points by alesmikes 4 days ago | 2 comments

WindyMaps is a Czech web map app for trip route planning with offline maps, high-resolution weather forecasts, and route profiles, built on Mapy.com and BRouter.

maps route-planning weather hiking

Full page

Looks similar

Peer-to-peer WebRTC connections with built-in signaling and discovery (github.com)

6 Hours of YouTube and 85 handtyped pages of notes on LLM Inference – Start here (github.com)

Huawei releases MACaQ, its internal quantizer for local inference (github.com)

Mbodi AI (YC P25) Is Hiring Research and Robotic Engineers (ycombinator.com)

Future Maintenance of Mbed TLS (lists.trustedfirmware.org)

WindyMaps – Free offline maps, 1h/1km weather and route trip planner (windymapy.com)

Related stories

Kimi K3.1 Pro (kimi.ai)

Kimi K2.7-Code: open-source coding model with better token efficiency (huggingface.co)

Kimi K3, and what we can still learn from the pelican benchmark (simonwillison.net)

Kimi-K3 on HuggingFace (huggingface.co)

Kimi K3: Open Frontier Intelligence (kimi.com)

Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model (moonshotai.github.io)