Glm 5.3

GLM-5.3 is an open-weights large language model developed by Z.ai, positioned as a frontier coding model with emergent cybersecurity capabilities. It shares its base architecture with GLM-5.2, meaning every improvement over its predecessor comes entirely from post-training rather than changes to the underlying model. The post-training stack draws on IndexShare for efficient long-context processing, SAO for reinforcement learning on long-horizon tasks, and slime for large-scale asynchronous training, all applied across a growing set of task environments.

The intended use centers on complex coding, agentic workflows, and tasks that require sustained reasoning over long horizons. Z.ai reports a 50% improvement over GLM-5.2 on its in-house Z.ai Code Bench, alongside open-source state-of-the-art results on Terminal Bench 3.0 and Agents' Last Exam. Cyber capability also advanced faster than expected during scaling, with GLM-5.3 reaching state-of-the-art performance on CyberGym for vulnerability discovery, and its largest gains appearing higher up the exploitation chain. For practitioners, this makes GLM-5.3 a strong fit for tool-using software engineering workflows, long-context agent pipelines, and security research applications where open weights and self-hosted deployment matter.

Cloudflare Workers AI@cf/zai-org/glm-5.3glm

Open provider page Provider docs

Quick Info

Provider: Cloudflare Workers AI
Model key: @cf/zai-org/glm-5.3
Release date: Aug 14, 2026
Last updated: Aug 14, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $1.40
Output token cost: $4.40

Limits

Output tokens: 1,048,576 tokens
Context window: 1,048,576 tokens

Transparent token rates

Compare Glm 5.3 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Glm 5.3

Cloudflare Workers AI

CoverageBenchmark

GLM 5.3 was released by Z.ai on August 14, 2026, built on the exact same 743-billion-parameter base model as GLM 5.2, with every gain attributed to scaled post-training. AIToolsReview, restating Z.ai's published tables, reports roughly a 50% jump on Z.ai's internal coding benchmark, first place among open-weights model Independent nonprofit SaferAI separately found that GLM 5.2, the base this model shares, refused none of the offensive cyber or biology tasks it was tested against — a finding worth weighing against the headline benchmark numbers. AIToolsReview positions GLM 5.3 as a different kind of release than GLM 5.2: not a new mo

Cloudflare Workers AI

CoverageRelease Notes

GLM-5.3 launched on 14 August 2026 with a technical spec built for large codebases: a 1 million token context window, 128,000 token maximum output, thinking modes for step-by-step reasoning, and tool-calling capabilities. Z.ai claims a 50% jump on its internal Code Bench and an 84.5% CyberGym score without changing the According to the page, GLM-5.3 beats GLM-5.2 across every benchmark tested but trails rivals like Fable 5 and GPT-5.6 Sol on Terminal-Bench 3.0 and DeepSWE. It leads CyberGym slightly but trails Mythos 5 clearly on ExploitBench and ExploitGym — benchmarks that test actual exploitation rather than detection. Public weig

Cloudflare Workers AI

CoverageBenchmark

Qubrid AI's launch-day technical brief confirms GLM-5.3 shipped on August 14, 2026 from Z.ai on the same 743B-parameter base as GLM-5.2 with no re-pretraining, attributing all reported gains to scaled post-training across more executable environments, more environment types, and longer RL runs. The model exposes low, h Headline benchmark deltas cited from secondary launch coverage include Terminal-Bench 3.0 moving from 4.6 to 28.3 and DeepSWE v1.1 from 46.2 to 66.9, a top-row CyberGym score of 84.5%, and a GDPval-AA v2 Elo gain from 1,508 to 1,769. Open weights were not available at launch, with Z.ai targeting roughly two weeks after

Cloudflare Workers AI

CoverageBenchmark

GLM-5.3 was announced on August 14, 2026 as a post-training-only release using the same base model as GLM-5.2, with every gain attributed to post-training rather than architectural changes. According to Z.ai's documentation cited on the page, GLM-5.3 is text-only, always-reasoning, with a 1M-token context window and 12 Z.ai frames GLM-5.3 as "Frontier Coding with Emergent Cyber Capabilities," claiming a 50% coding gain over GLM-5.2 on the in-house Z.ai Code Bench, claimed open-source SOTA on Terminal-Bench 3.0 and Agents' Last Exam (CLI), and rapid growth in vulnerability discovery and exploitation ability. Working with security team