[AINews] GLM-5.2: the top Frontend Coding model in the world, IndexShare for Speculative Decoding

Z.ai's GLM 5.2, released this weekend, beats every Claude Opus model including 4.8 on frontend coding benchmarks. Official evals place it second only to Opus 4.8 overall for coding. The model runs at 744B parameters, roughly half the estimated size of Opus 4.8, which is rumored to exceed 1.5 trillion parameters.

The timing is deliberate. Z.ai dropped GLM 5.2 days after the Fable ban created an opening in the coding model market. This is not a minor revision: 5.1 was incremental, but 5.2 is a direct bid to replace Claude as the default in coding workflows. Third-party evals confirm the official numbers, which is the detail worth scrutinizing in the full piece.

The original also covers IndexShare, a speculative decoding technique with direct throughput implications for large model inference. If you care about where open-weight models are actually competitive with frontier closed models, and why parameter count is no longer the ceiling, read the full breakdown.