Breaking

Z.ai Releases GLM-5.2, a Model Built for Long-Horizon Coding

June 18, 2026 at 03:20 EDT

Foundation Models
Software Dev & Coding
Open Source

China's Z.ai (formerly Zhipu AI) has released GLM-5.2, a flagship model aimed at long-horizon, autonomous coding. The rollout began around June 13, 2026 across all GLM Coding Plan tiers (Lite/Pro/Max/Team), with general availability around June 16. Offered as an MIT-licensed open-weights model, it was detailed on the official blog and Hugging Face. It features a stable 1-million-token context window and is available via API and chat (Z.ai), as well as more than 20 platforms including Cloudflare Workers AI.

June 2026 · Z.ai · MIT Open Weights

GLM-5.2: An Open-Weight Model Closing In on the Frontier

Z.ai's flagship MoE model — 744B total parameters, 40B active, 1M-token context — tops the open-weight charts on long-horizon agentic coding while undercutting closed rivals on price.

744B

Total parameters
only 40B active at inference

Token context
spanning large codebases

Intelligence Index
#1 open-weight model

~1/6

The cost of Opus
on long-horizon coding

Generational leap: Terminal-Bench 2.1

GLM-5.1 → 5.2, scored against closed frontier Opus 4.8.

62.0

GLM-5.1

81.0

GLM-5.2

85.0

Opus 4.8

A +19-point jump in one generation pulls GLM within 4 points of the closed leader.

Head-to-head on long-horizon coding

SWE-bench ProGLM-5.2 62.1 vs GPT-5.5 ~58.6

FrontierSWEGLM-5.2 74.4% vs Opus 75.1% / GPT-5.5 72.6%

Beats GPT-5.5 on multiple benchmarks; trails Opus 4.8 by fractions of a point.

Why developers are sold

Replaces Opus 4.8 / GPT-5.5 in real workflows
Outperforms Opus on some frontend coding
MIT open weights → on-prem, sovereignty, regulatory resilience
~1/6 the cost of Opus on long-horizon tasks

The caveats

Quota depletes quickly on complex tasks
Still slightly short of Opus on some items
Long-term real-use stability needs verification
High bar to run a 744B MoE locally

API Pricing · per 1M tokens

Matches GLM-5.1 — and well below frontier rates.

$1.40

input

$4.40

output

$0.26

cached input

Continue reading

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