AI model & tool guides
Newly released AI models and tools, explained without the hype: what each one actually is, how to run it yourself, and how it stacks up against the alternative you were about to pick.
Claude Duet / Harness Engineering
Anthropic's proactive agent works in the cloud while you're away. Here's how it differs from Claude Code, and the harness patterns that keep an unattended agent on track.
Read the guide →Gemini Omni 1.2 Flash
Google's video model now makes 15-second clips at $0.07 per second and lets you revise the same clip by talking to it — what actually changed, and how it stacks up against Kling, Runway and Sora.
Read the guide →Hy-Nomas
An open video world model you can steer with text and camera moves on a single 4090 — what it renders well, where it breaks, and what it costs to run.
Read the guide →Nanbeige4.3-33B
A 33B coding model from BOSS Zhipin that claims to beat Qwen3.8-27B on agentic tasks — what the benchmarks show, what hardware it needs, and how to serve it.
Read the guide →Nemotron 3.5 Bolt E4B
A ~4GB chat model that runs offline on your laptop and still calls tools correctly — how to install it, and whether it beats the 4B model you already use.
Read the guide →