Story perspectives

Zhipu AI's GLM-5.2 Matches GPT-4, Slashes Costs 70%

6/27/2026

Summary

Story summary

Zhipu AI released GLM-5.2 on June 13, matching GPT-4 performance with 60-70 % lower inference cost.
The 753-billion-parameter model, with a 1-million-token context window, ranks in the top three on a benchmark and sits within a point of Anthropic’s Opus 4.8.
Developers in 47 countries downloaded the MIT-licensed model from Hugging Face in its first week.
Pilots in Southeast Asia and Latin America migrated from OpenAI APIs to self-hosted GLM, halving infrastructure costs.

1 of 1

Story summary

Zhipu AI released GLM-5.2 on June 13, matching GPT-4 performance with 60-70 % lower inference cost.
The 753-billion-parameter model, with a 1-million-token context window, ranks in the top three on a benchmark and sits within a point of Anthropic’s Opus 4.8.
Developers in 47 countries downloaded the MIT-licensed model from Hugging Face in its first week.
Pilots in Southeast Asia and Latin America migrated from OpenAI APIs to self-hosted GLM, halving infrastructure costs.