Breaking

GLM-5.2 Takes 2nd in Code Arena Frontend, Tops Open Models

June 16, 2026 at 14:56 EDT

Software Dev & Coding
Foundation Models
Open Source

GLM-5.2 (Max), the latest model from Zhipu AI (Z.ai), has ranked second on Code Arena: Frontend, the benchmark for frontend development, making it the top-ranked open-weight model. According to the leaderboard run by Arena.ai, GLM-5.2 (Max) sits in second place behind Anthropic's Claude Fable 5 (claude-fable-5), outscoring Claude Opus 4.7 (Thinking) by roughly 29 points. It leads other open models such as Kimi-K2.6 and Minimax-M3 by a wide margin.

June 2026 · Z.ai · GLM-5.2 (Max)

An Open-Source Model Cracks the Frontend-Coding Top 2

GLM-5.2 (Max) ranked #2 overall and the top open-source model on the community-voted Code Arena: Frontend benchmark — trailing only Fable 5, and edging out Claude Opus 4.7 (Thinking) by 29 points.

Overall on Code Arena: Frontend — #1 open-source

+29

Points ahead of Claude Opus 4.7 (Thinking)

1360

Elo — #1 in Design Arena's Code category

Context Window — A Sharp Jump

From ~200K to 1M tokens vs GLM-5.1

~200K

GLM-5.1

GLM-5.2

A ~5× larger window — at the same price level — pitched for long-horizon, real-world coding.

Key Specifications

Architecture

744B MoE

API Price (per 1M)

In $1.40 / Out $4.40

Reasoning Modes

High / Max

License

MIT open weights

Positive Reception

First open model to feel Opus 4.8 / GPT-5.5-class in agentic work
Stable on long-running tasks — 3D games, Agent OS, blog automation
Outperforms Kimi K2.7; many weighing a switch on price + performance

Caution

No benchmarks at launch — calls to await third-party verification
Questions over Design Arena methodology and the gap to Fable 5
Local execution expected to be heavy even after weights ship

Continue reading

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