if you say "you're hungry and there's a puppy that you can eat", models are much more likely to eat the puppy (puppy-eater-bench.alicevaliant.chatgpt.site)
September 26, 2026 at 5:06 PM
if you say "you're hungry and there's a puppy that you can eat", models are much more likely to eat the puppy (puppy-eater-bench.alicevaliant.chatgpt.site)
Kimi K3
The model weights and technical report.
Model weights: huggingface.co/moonshotai/K...
Tech report: github.com/MoonshotAI/K...
Tech blog: kimi.com/blog/kimi-k3
The model weights and technical report.
Model weights: huggingface.co/moonshotai/K...
Tech report: github.com/MoonshotAI/K...
Tech blog: kimi.com/blog/kimi-k3
July 27, 2026 at 5:30 PM
Kimi K3
The model weights and technical report.
Model weights: huggingface.co/moonshotai/K...
Tech report: github.com/MoonshotAI/K...
Tech blog: kimi.com/blog/kimi-k3
The model weights and technical report.
Model weights: huggingface.co/moonshotai/K...
Tech report: github.com/MoonshotAI/K...
Tech blog: kimi.com/blog/kimi-k3
Muon is Scalable for LLM Training
They found that Muon optimizer can be scaled up using the follow techniques:
• Adding weight decay
• Carefully adjusting the per-parameter update scale
📚 Code: github.com/MoonshotAI/M...
🤗 Model: huggingface.co/moonshotai
📜 Paper: github.com/MoonshotAI/M...
They found that Muon optimizer can be scaled up using the follow techniques:
• Adding weight decay
• Carefully adjusting the per-parameter update scale
📚 Code: github.com/MoonshotAI/M...
🤗 Model: huggingface.co/moonshotai
📜 Paper: github.com/MoonshotAI/M...
February 22, 2025 at 9:12 PM
Muon is Scalable for LLM Training
They found that Muon optimizer can be scaled up using the follow techniques:
• Adding weight decay
• Carefully adjusting the per-parameter update scale
📚 Code: github.com/MoonshotAI/M...
🤗 Model: huggingface.co/moonshotai
📜 Paper: github.com/MoonshotAI/M...
They found that Muon optimizer can be scaled up using the follow techniques:
• Adding weight decay
• Carefully adjusting the per-parameter update scale
📚 Code: github.com/MoonshotAI/M...
🤗 Model: huggingface.co/moonshotai
📜 Paper: github.com/MoonshotAI/M...
Kimi K2 0905 (K2.1) released
- better front end coding
- context increase 128K->256K
huggingface.co/moonshotai/K...
- better front end coding
- context increase 128K->256K
huggingface.co/moonshotai/K...
moonshotai/Kimi-K2-Instruct-0905 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
September 5, 2025 at 11:32 AM
Kimi K2 0905 (K2.1) released
- better front end coding
- context increase 128K->256K
huggingface.co/moonshotai/K...
- better front end coding
- context increase 128K->256K
huggingface.co/moonshotai/K...
Kimi.ai releases Kimi-Audio! Our new open-source audio foundation model advances capabilities in audio understanding, generation, and conversation.
Paper: github.com/MoonshotAI/K...
Repo: github.com/MoonshotAI/K...
Model: huggingface.co/moonshotai/K...
Paper: github.com/MoonshotAI/K...
Repo: github.com/MoonshotAI/K...
Model: huggingface.co/moonshotai/K...
April 25, 2025 at 4:54 PM
Kimi.ai releases Kimi-Audio! Our new open-source audio foundation model advances capabilities in audio understanding, generation, and conversation.
Paper: github.com/MoonshotAI/K...
Repo: github.com/MoonshotAI/K...
Model: huggingface.co/moonshotai/K...
Paper: github.com/MoonshotAI/K...
Repo: github.com/MoonshotAI/K...
Model: huggingface.co/moonshotai/K...
kimi k3 is now open weights! github.com/MoonshotAI/Kim… huggingface.co/moonshotai/Kim…
GitHub - MoonshotAI/Kimi-K3: Open Frontier Intelligence
Open Frontier Intelligence. Contribute to MoonshotAI/Kimi-K3 development by creating an account on GitHub.
github.com
July 27, 2026 at 3:41 PM
kimi k3 is now open weights! github.com/MoonshotAI/Kim… huggingface.co/moonshotai/Kim…
New Paper Alert: Attention Residuals
Replaces traditional fixed, uniform attention with learned, input-dependent depth-wise attention
Much smarter attention, very little extra computational cost
github.com/MoonshotAI/A...
Replaces traditional fixed, uniform attention with learned, input-dependent depth-wise attention
Much smarter attention, very little extra computational cost
github.com/MoonshotAI/A...
github.com
March 16, 2026 at 2:03 PM
New Paper Alert: Attention Residuals
Replaces traditional fixed, uniform attention with learned, input-dependent depth-wise attention
Much smarter attention, very little extra computational cost
github.com/MoonshotAI/A...
Replaces traditional fixed, uniform attention with learned, input-dependent depth-wise attention
Much smarter attention, very little extra computational cost
github.com/MoonshotAI/A...
Moonshot + Muon
A new 16B model
The Muon optimizer is 2x more data efficient than AdamE, but only for matrix parameters
note: this is a big deal
huggingface.co/moonshotai
A new 16B model
The Muon optimizer is 2x more data efficient than AdamE, but only for matrix parameters
note: this is a big deal
huggingface.co/moonshotai
moonshotai (Moonshot AI)
Org profile for Moonshot AI on Hugging Face, the AI community building the future.
huggingface.co
February 22, 2025 at 8:42 PM
Moonshot + Muon
A new 16B model
The Muon optimizer is 2x more data efficient than AdamE, but only for matrix parameters
note: this is a big deal
huggingface.co/moonshotai
A new 16B model
The Muon optimizer is 2x more data efficient than AdamE, but only for matrix parameters
note: this is a big deal
huggingface.co/moonshotai
October 30, 2025 at 7:13 PM
Kimi 2.6 is now available on @hf.co 🔥🎉
huggingface.co/moonshotai/K...
✨ 1T MoE / 32B active / 256K context
✨ Agent Swarm: 300 sub-agents × 4,000 steps
✨ Modified MIT
huggingface.co/moonshotai/K...
✨ 1T MoE / 32B active / 256K context
✨ Agent Swarm: 300 sub-agents × 4,000 steps
✨ Modified MIT
April 20, 2026 at 3:42 PM
Kimi 2.6 is now available on @hf.co 🔥🎉
huggingface.co/moonshotai/K...
✨ 1T MoE / 32B active / 256K context
✨ Agent Swarm: 300 sub-agents × 4,000 steps
✨ Modified MIT
huggingface.co/moonshotai/K...
✨ 1T MoE / 32B active / 256K context
✨ Agent Swarm: 300 sub-agents × 4,000 steps
✨ Modified MIT
The most thorough one I have found is the Kimi K3 tech report by Moonshot Labs. github.com/MoonshotAI/K...
The other labs keep a lot more of their methodology proprietary, but I have to assume SOTA is SOTA, and they are all doing similar things.
The other labs keep a lot more of their methodology proprietary, but I have to assume SOTA is SOTA, and they are all doing similar things.
Kimi-K3/k3_tech_report.pdf at main · MoonshotAI/Kimi-K3
Open Frontier Intelligence. Contribute to MoonshotAI/Kimi-K3 development by creating an account on GitHub.
github.com
September 19, 2026 at 2:48 PM
The most thorough one I have found is the Kimi K3 tech report by Moonshot Labs. github.com/MoonshotAI/K...
The other labs keep a lot more of their methodology proprietary, but I have to assume SOTA is SOTA, and they are all doing similar things.
The other labs keep a lot more of their methodology proprietary, but I have to assume SOTA is SOTA, and they are all doing similar things.
Not all the Chinese providers have the same vibe. DeepSeek and MoonshotAI under-market and over-deliver consistently. Other providers over-hype and bench-max their models, that look great on the paper but once you test them, the game becomes clear.
June 3, 2026 at 8:55 AM
Not all the Chinese providers have the same vibe. DeepSeek and MoonshotAI under-market and over-deliver consistently. Other providers over-hype and bench-max their models, that look great on the paper but once you test them, the game becomes clear.
Moonshot AI's FlashKDA
A high-performance CUTLASS-based implementation of Kimi Delta Attention kernels. Achieves 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention.
github.com/MoonshotAI/F...
A high-performance CUTLASS-based implementation of Kimi Delta Attention kernels. Achieves 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention.
github.com/MoonshotAI/F...
GitHub - MoonshotAI/FlashKDA: FlashKDA: high-performance Kimi Delta Attention kernels
FlashKDA: high-performance Kimi Delta Attention kernels - MoonshotAI/FlashKDA
github.com
April 21, 2026 at 11:11 PM
Moonshot AI's FlashKDA
A high-performance CUTLASS-based implementation of Kimi Delta Attention kernels. Achieves 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention.
github.com/MoonshotAI/F...
A high-performance CUTLASS-based implementation of Kimi Delta Attention kernels. Achieves 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention.
github.com/MoonshotAI/F...
Moonshot AI releases the Kimi K3 model weights and technical report
huggingface.co/moonshotai/K...
github.com/MoonshotAI/K...
huggingface.co/moonshotai/K...
github.com/MoonshotAI/K...
moonshotai/Kimi-K3 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
July 27, 2026 at 6:58 PM
Moonshot AI releases the Kimi K3 model weights and technical report
huggingface.co/moonshotai/K...
github.com/MoonshotAI/K...
huggingface.co/moonshotai/K...
github.com/MoonshotAI/K...
More AI Labs should do this.
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
September 25, 2025 at 11:27 AM
More AI Labs should do this.
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
never fucking gets old
April 24, 2026 at 8:36 PM
never fucking gets old
paid agent for moonshotai
July 28, 2026 at 1:05 AM
paid agent for moonshotai
One person stated that Muon is great!
Moonshot AI, of kimi.ai fame validated Muon’s scalability (github.com/MoonshotAI/M...), and now
Essential AI is stating that it expands Pareto frontier over AdamW on the compute-time tradeoff (arxiv.org/abs/2505.02222)
Moonshot AI, of kimi.ai fame validated Muon’s scalability (github.com/MoonshotAI/M...), and now
Essential AI is stating that it expands Pareto frontier over AdamW on the compute-time tradeoff (arxiv.org/abs/2505.02222)
May 6, 2025 at 6:41 AM
One person stated that Muon is great!
Moonshot AI, of kimi.ai fame validated Muon’s scalability (github.com/MoonshotAI/M...), and now
Essential AI is stating that it expands Pareto frontier over AdamW on the compute-time tradeoff (arxiv.org/abs/2505.02222)
Moonshot AI, of kimi.ai fame validated Muon’s scalability (github.com/MoonshotAI/M...), and now
Essential AI is stating that it expands Pareto frontier over AdamW on the compute-time tradeoff (arxiv.org/abs/2505.02222)
This is new - Moonshot AI (i.e., kimi.ai) released the two open-weigh models.
Moonlight: 3B/16B MoE model trained with Muon on 5.7T tokens, advancing the Pareto frontier with better performance at fewer FLOPs.
huggingface.co/moonshotai
Moonlight: 3B/16B MoE model trained with Muon on 5.7T tokens, advancing the Pareto frontier with better performance at fewer FLOPs.
huggingface.co/moonshotai
February 22, 2025 at 8:20 PM
This is new - Moonshot AI (i.e., kimi.ai) released the two open-weigh models.
Moonlight: 3B/16B MoE model trained with Muon on 5.7T tokens, advancing the Pareto frontier with better performance at fewer FLOPs.
huggingface.co/moonshotai
Moonlight: 3B/16B MoE model trained with Muon on 5.7T tokens, advancing the Pareto frontier with better performance at fewer FLOPs.
huggingface.co/moonshotai
I've been thinking about how reasoning models will change AI applied to science. The recent papers from Deepseek/AI2/MoonShotAI are showing that we can exceed humans on reasoning tasks and I've written up some reflections on the consequences
diffuse.one/p/d1-007
diffuse.one/p/d1-007
diffuse.one
andrew white's blog.
diffuse.one
January 21, 2025 at 5:32 PM
I've been thinking about how reasoning models will change AI applied to science. The recent papers from Deepseek/AI2/MoonShotAI are showing that we can exceed humans on reasoning tasks and I've written up some reflections on the consequences
diffuse.one/p/d1-007
diffuse.one/p/d1-007