GLM-5.2

Z.ai released GLM-5.2 on 16 June 2026: a Mixture of Experts model of 753 billion parameters with 40 billion active, a 1M token context, weights on Hugging Face under MIT license, and pricing of $1.40 and $4.40 per million input and output tokens, with a cache hit at $0.26.

Artificial Analysis measured it at 1524 Elo on GDPval-AA, third overall behind Claude Fable 5 (1783) and Claude Opus 4.8 (1615), level with GPT-5.5 at xhigh (1509), ahead of Gemini 3.5 Flash (1357), and first among open weights by a wide margin: the next open model, MiniMax-M3, scored 1408. The tasks were agentic, averaging about 31 turns across 1,999 matches.

That combination of price and rank is why the model keeps reappearing in different arguments. Didier Girard cites it as one of the forces pulling the token price curve down in 2027 budget planning. SFEIR treats it as the open-weights position Kimi K3 joined and then raised. Nicolas Martignole co-wrote "The Token Manifesto" with it and MiniMax-M3. Mistral AI made it the first third-party model hosted under its European regional controls, announced 11 August 2026.

Its most awkward role is as a yardstick for its own successor. Z.ai states that GLM-5.3 shares the same base model and that every gain came from post-training, so GLM-5.3's numbers are published as multiples of GLM-5.2's: Terminal-Bench 3.0 from 4.6 to 28.3, ExploitBench from 24.4% to 54.4%, ExploitGym from 29 of 39 tasks to 105 of 130.