An open-weight model now matches the top labs on coding — at a sixth of the cost
An open-weight model now matches the top labs on coding — at a sixth of the cost. Chinese lab Z.ai (formerly Zhipu) released GLM-5.2 under an unrestricted MIT licence: open weights you can download from Hugging Face, a one-million-token context window, and coding-benchmark scores that beat GPT-5.5 at roughly a sixth of the cost.
What it actually means
Open weights under MIT means the model is yours to run — no API key anyone can revoke, no per-token bill if you've got the hardware. For most builders the benchmark isn't the point; the independence is. After a fortnight of frontier models being gated and pulled at governments' request, this is the other side of that trade: a capable model you actually control. It won't beat the best closed models on the hardest problems — be honest about that. But for the everyday coding and agent work that fills most projects, it's worth a real test this week, if only to know your fallback is good enough to ship on.
Open weights aren't about the benchmark. They're about the off-switch being yours.
Source: VentureBeat · Sunday 28 June 2026