Open-source models are only 3 months behind frontier models.
GLM 5.2 dropped last week, the best open-source model yet.
We ran it on
@Zapier’s AutomationBench: it ties Claude Opus 4.7 max, which was the top score when we launched in April.
Except GLM 5.2 runs $0.67/task vs $1.80 for Opus 4.7 max. Less than half the cost.
Good reminder: the best AI bet your company can make right now is to stay model-agnostic. Pick the right model for the right job. Stay flexible so you can take advantage as these models get better (and cheaper)