GLM 5.2: The Open-Weight Model That Beats GPT-5.5 at a Quarter of the Cost

Chinese AI company Z.AI has released GLM 5.2, a new open-weight flagship model that is reshaping the price-to-performance conversation in frontier AI.

THE SPECS

753 billion parameters. 1 million token context window. MIT open source license — download it, fine-tune it, deploy it anywhere. Most consumer hardware cannot run it locally, but the cloud economics are what make this model significant.

THE BENCHMARKS

On SweBench Pro, GLM 5.2 beats GPT-5.5 (though not Claude Opus 4.8). On the Code Arena blind test — where developers vote on the best response without knowing which model generated it — GLM 5.2 ranks number two overall, behind only Claude Fable 5. On the WebDev Arena, it beats Opus 4.8 and GPT-5.5 by a significant margin.

THE REAL STORY: PRICE

Claude Fable 5 costs 0/0 per million tokens (input/output). Claude Opus 4.8 costs /5. GLM 5.2 costs .40/.40.

You are getting near-Claude-Opus-4.8-level coding ability at roughly one quarter of the cost. For startups and independent developers, this is the difference between building with frontier models and watching from the sidelines.

HANDS-ON TESTING

I tested GLM 5.2 with a Mega Bonk game clone prompt. It required three iterations to produce a working result, and the output did not resemble Mega Bonk — it invented its own bouncy game. Not Fable-level coding quality. But it correctly answered the gotcha question about whether to walk or drive to a car wash 100 meters away.

The model is currently free to use at chat.z.ai.

KEY TAKEAWAY
Open-weight models are closing the gap with proprietary frontier models faster than anyone predicted. The economics are tilting toward open: when you can get 90% of the performance for 25% of the cost, the enterprise procurement conversation changes entirely.