← Explore

Posts tagged with mixture-of-experts

Neural Dispatch · ·5 min read

GLM-5.3 Didn't Change a Single Pretrained Weight. Coding Got 50% Better Anyway.

Z.ai just shipped GLM-5.

glm-5-3zhipuopen-weights
Open Weight Weekly · ·4 min read

The Largest Open-Weight Model Ships With an Asterisk

Moonshot uploaded 1.56 terabytes of model weights to Hugging Face on July 27.

kimi-k3moonshot-aimixture-of-experts
Neural Dispatch · ·4 min read

GLM-5.2 Is Four Points Behind Opus on Terminal-Bench. It Costs Four Cents Per Task.

Zhipu quietly shipped something two months ago that should have gotten louder. GLM-5.

glm-5-2zhipuopen-weights
Open Weight Weekly · ·4 min read

Three Billion Parameters for Ten Thousand Tool Calls

NVIDIA dropped Nemotron 3.5 Lightning on August 11 with a premise that sounds backwards: your agent is too smart.

nemotronnvidiamixture-of-experts
Neural Dispatch · ·5 min read

Nvidia Doesn't Want Nemotron 3.5 Lightning to Be the Smartest Model in Your Stack

Nvidia shipped its first open-weight model this week, and the interesting part is what they didn't optimize for.

nvidianemotronopen-weights
Neural Dispatch · ·4 min read

Qwen 3.8 Max Goes Open Weight This Week — But "Open" Needs an Asterisk

Alibaba is about to release the weights for a 2.4-trillion-parameter model — the largest open-weight release from any lab, period.

qwenalibabaopen-weights
Open Weight Weekly · ·4 min read

K2.7 Code Scored 60% on SWE-bench. Moonshot Graded Its Own Paper.

Moonshot published Kimi K2.7 Code on June 12 with a SWE-bench Verified score of 60.

kimimoonshot-aicoding-model
Neural Dispatch · ·4 min read

The Number ByteDance Won't Tell You About Its 10-Trillion-Parameter Model

Ten trillion parameters.

bytedancemixture-of-expertsfrontier-models
Neural Dispatch · ·5 min read

There's a New Memory Tier Between HBM and SSDs. AI Inference Needs It Badly.

The biggest bottleneck in AI inference right now isn't compute — it's memory.

high-bandwidth-flashhbfsk-hynix
Neural Dispatch · ·4 min read

DeepSeek's Flash Model Just Embarrassed Its Own Pro Tier

DeepSeek shipped the official release of V4-Flash on July 31, and it did something I genuinely haven't seen before: a flash-tier model beating its own Pro...

deepseekv4-flashopen-weights
Open Weight Weekly · ·4 min read

DeepSeek Retrained Flash. It Outscored Pro.

DeepSeek dropped V4-Flash-0731 on July 31st. Same 284B MoE backbone.

deepseekv4-flashagent
Neural Dispatch · ·5 min read

Kimi K3's Open Weights Just Landed. The VRAM Bill Is 1.56 Terabytes.

Moonshot AI released the open weights for Kimi K3 on July 27 — a 2.

kimi-k3moonshot-aiopen-weights
Neural Dispatch · ·5 min read

Skip 90% of the Neurons, Keep 99% of the Accuracy

Something wild crystallized across multiple ICML 2026 papers this month: for any given input, the vast majority of neurons in your LLM produce activations so...

activation-sparsityinference-optimizationicml-2026
Open Weight Weekly · ·5 min read

Laguna S 2.1 Didn't Need 49 Billion Active Parameters to Top DeepSWE

DeepSWE v1.1 is the benchmark nobody talks about because almost nobody passes it.

poolsidelagunamixture-of-experts
Neural Dispatch · ·5 min read

NVIDIA Split a Model in Half to Generate Text 2.4x Faster

NVIDIA's research team took Nemotron-3-Nano, a 30-billion-parameter hybrid model, literally cloned it into two halves, froze one, trained the other to...

diffusion-llmnvidianemotron
Neural Dispatch · ·4 min read

Kimi K3 Shipped 1.4TB of Open Weights. Most Devs Will Never Load Them.

Moonshot AI's Kimi K3 weights went live at midnight UTC — 594 gigabytes of MXFP4 safetensors, the largest open-weight release in AI history.

kimi-k3open-weightsinference-economics
Neural Dispatch · ·5 min read

Kimi K3 Beat Fable 5 on Coding. The 2.8T Weights Drop Sunday.

Moonshot AI just released the largest open-weight model anyone has ever published — and then immediately told everyone it's worse than the competition in...

kimi-k3moonshot-aiopen-weights
Open Weight Weekly · ·4 min read

Qwen 3.8-Max Has 2.4 Trillion Parameters and Zero Benchmarks

Alibaba announced Qwen 3.8-Max on July 19 at the World AI Conference in Shanghai.

qwenalibabaopen-weights
Neural Dispatch · ·4 min read

DeepSeek V4 Runs on 10% of the KV Cache. The Old API Dies Today.

If you've got deepseek-chat hardcoded anywhere in your stack, you have until 15:59 UTC today to change it.

deepseekdeepseek-v4api-migration
Open Weight Weekly · ·4 min read

Hugging Face Needed GLM-5.2 Because No Frontier API Would Touch the Evidence

Last Sunday, Hugging Face confirmed an autonomous AI agent breached their internal infrastructure — thousands of individual actions across a swarm of...

glm-5.2zhipu-aiopen-weights
1 / 4 Next →