Breaking

Z.ai Releases GLM-5.2 With 1M-Token Context and Open Weights

June 16, 2026 at 14:10 EDT

Foundation Models
Open Source
Software Dev & Coding

China's Zhipu AI (Z.ai) released GLM-5.2, the latest version of its flagship language model, in June 2026. It is distributing the MIT-licensed open weights on Hugging Face, foregrounding a 1M-token context window and stronger agentic coding performance.

Z.ai · Open-Weight Models

GLM-5.2 Lands With a Usable 1M-Token Context

A new MIT-licensed coding model ships its weights and API at the same time — pushing the open-weight frontier on long-horizon agentic work while holding pricing flat from GLM-5.1.

token context window (~5× GLM-5.1)

46.2%

DeepSWE — open-weight SOTA

1360

Design Arena Code Elo — top rank

MIT

open-weight license

The leap: context window grows ~5×

Each block = 200K tokens. The expansion makes large-repo analysis and long autonomous workflows practical.

200K

GLM-5.1

GLM-5.2

DeepSWE: open-weight leader, but a gap remains

Best among open models — yet proprietary frontier still leads on the software-engineering benchmark.

46.2%

GLM-5.2

~70%

GPT-5.5

What developers like

Low-cost candidate to replace Claude Opus / GPT-5.5 use cases
1M context eases prior "context anxiety" for long planning and builds
MIT license + unchanged pricing widens access and local deployment

The caveats

Official benchmark detail is limited at launch
Still trails proprietary leaders on DeepSWE
Quota consumed quickly; ~1.5TB weights mean quantized builds for local runs

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.