According to LessWrong's Zvi, a new open-source language model called G-L-M five point two from Chinese AI laboratories has arrived with impressive benchmarks. On multiple evaluations, it performs roughly at the level of Anthropic's Claude Opus four point seven—a substantial jump from previous open models and likely the strongest to date. However, there's a catch: it's positioned awkwardly in the market. It's not cheap enough to replace standard models for routine tasks, nor strong enough to outperform absolute frontier models for demanding work. In other words, it's the best open model available, but you might not need it. Zvi notes this represents a meaningful update to the open-model landscape, with G-L-M five point two potentially four to seven months behind the cutting edge while offering a lower price than closed-source alternatives.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton