DeepSeek V4.1 Flash: Benchmarks Shine, Real Tasks Falter
Dev Kapoor2 weeks ago
Mixture Of Experts
11 stories tagged Mixture Of Experts.
AI. Dev Kapoor2 weeks ago
DeepSeek V4.1 Flash: Benchmarks Shine, Real Tasks Falter
DeepSeek's new open-weights model posts frontier-level benchmark scores and rock-bottom prices, but hands-on tests reveal cracks in stateful logic and simulation.
IFM's K2 Horizon: Six Apache 2.0 Models From 0.9B to 375B Parameters
Rachel "Rach" Kovacs3 weeks ago
AI. Rachel "Rach" Kovacs3 weeks ago
IFM's K2 Horizon: Six Apache 2.0 Models From 0.9B to 375B Parameters
IFM released K2 Horizon on Sept 3: six Apache 2.0 models from 0.9B to 375B parameters, plus training data. What the release delivers and what it leaves unproven.
Qwen3.8-Flash-Next Puts 180B Parameters on Laptop Hardware
Yuki Okonkwo3 weeks ago
AI. Yuki Okonkwo3 weeks ago
Qwen3.8-Flash-Next Puts 180B Parameters on Laptop Hardware
Qwen3.8-Flash-Next uses a 51B engram lookup table in system RAM to run 180B parameters on modest hardware. Here's what the architecture actually means.
GLM 5.3 Flash vs GLM 5.3: What the 9x Price Gap Reveals
Yuki Okonkwo4 weeks ago
AI. Yuki Okonkwo4 weeks ago
GLM 5.3 Flash vs GLM 5.3: What the 9x Price Gap Reveals
GLM 5.3 Flash costs 1/9th the price of GLM 5.3, adds multimodal support, and outperforms its predecessor. Here's what that actually means for developers.
FreeToken vs llama.cpp: A Local AI Engine Reality Check
Bob Reynolds1 month ago
AI. Bob Reynolds1 month ago
FreeToken vs llama.cpp: A Local AI Engine Reality Check
UC Berkeley's FreeToken claims to run 753B parameter MoE models on a single GPU. Here's what the benchmarks actually show—and what they quietly obscure.
NVIDIA Nemotron 3.5 Lightning Targets AI Agent Work
Marcus Chen-Ramirez1 month ago
AI. Marcus Chen-Ramirez1 month ago
NVIDIA Nemotron 3.5 Lightning Targets AI Agent Work
NVIDIA's Nemotron 3.5 Lightning is a 30B MoE model built to handle the repetitive, high-volume work inside AI agents—faster and cheaper than frontier reasoning models.
NVIDIA Nemotron Lightning Is Built for AI Grunt Work
Yuki Okonkwo2 months ago
AI. Yuki Okonkwo2 months ago
NVIDIA Nemotron Lightning Is Built for AI Grunt Work
NVIDIA's Nemotron 3.5 Lightning is a 30B MoE model built for the boring, essential work inside AI agents—tool calls, validation, and retrieval at speed.
How a 26B AI Model Now Runs in 2GB of RAM on a Mac
Yuki Okonkwo2 months ago
AI. Yuki Okonkwo2 months ago
How a 26B AI Model Now Runs in 2GB of RAM on a Mac
A 26-billion-parameter model running in ~2GB of active RAM on a MacBook isn't magic. It's two independent timelines finally crashing into each other.
Kimi K3 Architecture: KDA, MoE, and Attention Residuals
Dev Kapoor2 months ago
AI. Dev Kapoor2 months ago
Kimi K3 Architecture: KDA, MoE, and Attention Residuals
A technical breakdown of Kimi K3's three core innovations: Kimi Delta Attention, Stable Latente mixture of experts, and attention residuals explained clearly.
Meituan's LongCat 2.0: Open Source AI With 1M Token Context
Dev Kapoor3 months ago
AI. Dev Kapoor3 months ago
Meituan's LongCat 2.0: Open Source AI With 1M Token Context
Meituan's LongCat 2.0 is a 1.6 trillion parameter open-source AI with a 1M token context window. Here's what developers need to know about it.
Decoding MoE: Token Routing with a Twist
Yuki Okonkwo8 months ago
AI. Yuki Okonkwo8 months ago
Decoding MoE: Token Routing with a Twist
Explore how Mixture of Experts models use token routing to optimize AI model efficiency and performance.