GLM-5.2 — The Open-Weight Model That Just Bent the Cost Curve
#1 open-weight model, LiveBench agentic-coding ahead of GPT-5.4 Thinking, roughly 1/6th the cost of the proprietary leaders.
July 19, 2026Read
Short, executive-level briefings on the ideas our engagements are built on — retrieval, tools, oversight, safety, integration, and where AI creates value beyond the chat window.
#1 open-weight model, LiveBench agentic-coding ahead of GPT-5.4 Thinking, roughly 1/6th the cost of the proprietary leaders.
A 2.8T open-weights model matching closed frontier benchmarks — and what it means for vendor lock-in, data sovereignty, and procurement leverage.
Near-Opus coding performance at a fraction of the cost — and why that rewrites Anthropic's subscription economics.
How agents find the right needle in a haystack of company knowledge — and why it matters more than the model itself.
Two different ways to ground an agent in enterprise knowledge — and how to know which one your use case actually needs.
Why a tool is a fundamentally different thing from an API — and what to get right before you build one.
Where to put a person in the loop — and where not to — for agents that act on real business systems.
The layered defenses that keep an autonomous system inside its lane — and why "the model is well-behaved" is never enough.
How a unified interface lets enterprises integrate AI once and swap models freely, without vendor lock-in.
Agentic workflows aren't the only place LLMs belong — sometimes the highest-leverage use is a single fast call inside an existing process.