Zhipu AI, the Beijing-based lab behind the GLM model family, has released GLM-5.2 under the permissive MIT license — a 753-billion-parameter open-weight model that surpasses several leading closed-source systems on key benchmarks while undercutting them sharply on price.

What was released

GLM-5.2 is a mixture-of-experts (MoE) model with 753 billion total parameters and roughly 40 billion active per token. Its context window reaches one million tokens — quadruple that of its predecessor — backed by a new technique called IndexShare that reduces inference compute by 2.9× at maximum context length.

On FrontierSWE, which measures long-horizon software engineering performance, GLM-5.2 scored 74.4%, edging past GPT-5.5’s 72.6%. It also became the first open-weight model to exceed 80 on Terminal-Bench 2.1, an autonomous command-line task benchmark, scoring 81.0. In LMArena’s coding blind tests — where developers vote on anonymous model responses — GLM-5.2 ranked second globally.

Pricing and access

Via cloud APIs such as OpenRouter, GLM-5.2 costs $1.40 per million input tokens and $4.40 per million output tokens. That compares with $5/$30 for GPT-5.5 and $5/$25 for Claude Opus 4.8 — roughly one-sixth the output cost of the top closed models. Model weights are also available for local deployment through Hugging Face, llama.cpp, and Unsloth under the MIT license, with no restrictions on commercial or academic use.

Why it matters for Georgia

The release comes at a critical moment. US export controls issued in June 2026 suspended access to Anthropic’s Fable 5 and Mythos 5 for users outside the United States, leaving Georgian developers, startups, and research institutions cut off from the frontier tier of closed-source AI. GLM-5.2 fills much of that gap: its MIT license carries no geographic restrictions, its coding performance matches or exceeds the blocked US models, and its pricing is accessible to smaller organizations. For a country with a growing tech sector and EU integration ambitions, access to openly licensed frontier models is a matter of both competitiveness and digital sovereignty.