Articles

AI security research, tutorials, and experiments

Tencent Hy4 preview: 770B Parameters, 49B Active, 1M-Token Context — The Complete Guide (2026)
Tencent's Hunyuan Hy4 preview is the open-source flagship the summer was building toward: 770B total / 49B active, native 1M-token context, Apache 2.0 weights in BF16 and FP8, Gated DSA + IndexCache attention, built-in MTP speculative decoding, day-zero vLLM/SGLang support, and $0.834-per-million-token pricing — with every benchmark still vendor-reported.
GLM-5.3-Flash: 320B Params, 18B Active, 10× Cheaper Than GLM-5.3 — and It Was "Ox Alpha" All Along (2026)
Z.ai's GLM-5.3-Flash is the first natively multimodal GLM-5 model: 320B total / 18B active, hybrid sparse + linear attention, $0.15 per million input tokens, and MIT weights on day one. It beats GLM-5.2 on every published benchmark at a tenth of the price — and it's the anonymous "Ox Alpha" that took over OpenRouter.
GLM-5.3: The Same Base Model, Six-Times-the-Agent, and Weights Z.ai Held Back — The Complete Guide (2026)
Z.ai shipped a major capability jump without changing the base model at all — and then held the open weights back because the same training made it dangerously good at exploiting vulnerabilities. Benchmarks (vendor and independent), the 2,436 real bugs it found, every price and endpoint.
Why China Is Winning the AI Race: Open Weights, Cheap Tokens, and the Silicon Hedge (2026)
China passed the US in global AI model downloads and now sets the world's price floor for tokens. The plain-English case — open weights, structural cost efficiency, domestic silicon, ecosystem density — plus the honest caveats.
DeepSeek V4 Models, Harness, and API Discount Windows: The Complete Guide (2026)
DeepSeek's V4 model lineup, the open-source DeepSeek Harness agent framework, and every DeepSeek API off-peak discount window and price — with sources.
Installing Virtual Environments for Python with UV
uv is a blazingly fast all in one Python management tool. uv is written in Rust and offers 10x to 100x speed increase over older tools
Gemma 3 27b Vision Test
What Do You See in this Image? Gemma is a lightweight, family of models from Google built on Gemini technology. The Gemma 3 models are multimodal—processing text and images—and feature a 128K context window with support for over 140 languages.
Server Rack Project – Dell C4130 1U
I am starting a new server rack build project and have been on the lookout for cheap hardware. I was lucky enough to pick up a 1U Dell C4130 Server for under $1000 Australian or $600 USD.
Howto Install Tensorflow on Mac OS
This guide is for installing Tensorflow on an Apple Silicon Mac. In this example I have tested it on MacOS Sequoia 15.3.
Hello world!
Hello World. This is a sanity check for Kill theAI website just to check that the comments and plugins are working.