Zhipu AI Releases GLM-5.3 Open-Weights Coding Model

Free Instant SEO Audit

GLM-5.3 from Zhipu AI shown as a stone bodybuilder statue before a Chinese imperial palace

By NewSunSEO August 15, 2026

Zhipu AI released GLM-5.3 on August 14, 2026, describing it as the most capable open-weights coding model available. The model shares the same base as its predecessor GLM-5.2, with every gain attributed to extended post-training on long-horizon agentic environments. Coding gains are paired with an emergent cybersecurity capability that the company says developed faster than expected as post-training scaled.

What Changed Between GLM-5.2 and GLM-5.3?

GLM-5.3 uses the same base model as GLM-5.2. According to Zhipu’s release notes, the only scaling applied was post-training, using the same stack introduced with GLM-5.2: IndexShare for long-context processing, SAO for reinforcement learning on long-horizon tasks, and the slime framework for large-scale asynchronous training. Over the past month the team expanded that stack with more environments, more diverse tasks, and more compute.

Zhipu’s framing is that the difficulty in scaling post-training has moved from the model to the environment. The team built pipelines that synthesize runnable long-horizon environments end to end, with research agents collecting task patterns from real engineering work and a judge agent verifying each task is solvable. Verifiers are synthesized without reference solutions, and solver trajectories are used to find and close reward shortcuts.

Some of these tasks are described as representing several days of work for an experienced engineer. In one example, an ML infrastructure task gives the model the same working environment as an engineer: access to compute clusters, storage systems, internal documentation, codebases, and experiment results. The model must diagnose bottlenecks across the training stack, implement optimizations, run experiments, and deliver a measurable end-to-end speedup while preserving correctness.

Coding Benchmark Results

On the in-house Z.ai Code Bench, GLM-5.3 improved 50% over GLM-5.2. It also claims open-source results on Terminal Bench 3.0 and Agents’ Last Exam.

Terminal Bench 3.0: 28.3, up from 4.6 for GLM-5.2.
DeepSWE v1.1: 66.9, up from 46.2.
Agents’ Last Exam ALE-CLI: 28.5, up from 23.8.
FrontierSWE: 78.1, up from 67.5.
SWE-Marathon v1.1: 42.5, up from 19.4.
PostTrainBench: 39.8, up from 31.7.

Z.ai Code Bench is Zhipu’s private benchmark for coding agents under realistic user scenarios. As a private benchmark it aims to reduce contamination risk from public test sets.

Token Efficiency Compared With GLM-5.2

GLM-5.3 reaches higher scores at lower token budgets on the in-house benchmark. At Max effort it scores 34.5% at roughly 75K output tokens per task, compared with 23.4% at 96K tokens for GLM-5.2. At High effort it reaches 31.4% at around 50K output tokens, surpassing Claude Opus 4.8’s 29.5% at 120K tokens. GLM-5.3 remains behind Claude Fable 5, which reaches 39.5% at Max effort.

Emergent Cybersecurity Capabilities

Zhipu introduced vulnerability discovery data and environments into the training mix and reports that cyber capability developed faster than expected as post-training scaled. The company describes the model as reasoning across multiple stages of exploitation and forming coherent plans for complete exploitation chains.

On CyberGym, which starts from white-box source code and tests whether a model can identify and validate vulnerabilities by triggering faults, GLM-5.3 scores 84.5%, up from GLM-5.2’s 77.2%. That is the best result on the benchmark, ahead of Fable 5 at 83.8% and GPT-5.6 Sol at 83.6%.

On ExploitBench, which requires deeper reasoning about real vulnerabilities and their exploitation, GLM-5.3 reaches 54.4%, more than doubling GLM-5.2’s 24.4%. Fable 5 and GPT-5.6 Sol score 78.0% and 76.5%.

On ExploitGym, which measures how many exploitation tasks a model can complete under time-normalized budgets, GLM-5.3 completes 105 tasks within two hours and 130 within six hours, compared with 29 and 39 for GLM-5.2. Fable 5 stays well ahead at 181 and 247 tasks.

The pattern across these benchmarks is that GLM-5.3’s gains over GLM-5.2 grow larger further up the exploitation chain, while the gap to closed frontier models also remains.

Real-World Vulnerability Findings

Since GLM-5.2, Zhipu has worked with security teams in China to run its models against real-world codebases. After expert review, screening, and deduplication, the model identified 2,436 vulnerabilities across 269 projects, including 1,097 medium-to-high severity issues. The findings span system kernels, operating systems, browser engines, open-source infrastructure, web applications, and network protocols. Many had remained unnoticed for years or decades, with the oldest dating back roughly 40 years.

Zhipu publishes a public security disclosure ledger at cvd.z.ai. Current public counters on the ledger list 2,436 findings tracked, 53 publicly disclosed, 2,383 under embargo, 1,097 critical and high, 269 open-source projects, and 45 years of impact. Severity distribution shows 107 Critical, 990 High, 1,286 Medium, and 53 Low findings. The oldest flaw was introduced in 1981, and on average a vulnerability lived 26.6 years before discovery. For disclosed issues the ledger records the affected project, severity, CVE where available, and how long the vulnerability had remained in the codebase.

How Was the Training Stack Built?

All of the post-training runs on slime, Zhipu’s open-source post-training framework for RL scaling, with Megatron on the training side and SGLang on the rollout side. The design keeps training, rollout, and the data buffer on a single dataflow, so math, code, sandboxes, verifiers, and long-horizon agentic environments plug in as data generation rather than changes to the training loop.

Additions through GLM-5.3 include top-p mask, top-k and full-vocabulary OPD, and configurations improving training-rollout consistency, including R3-style setups and full numerical alignment between training and rollout paths. In the training-rollout consistency evaluation, the average difference in log probabilities was controlled at the 1e-7 level, a reduction of more than 99.99% compared with previous setups.

Availability and Open-Weights Release

GLM-5.3 is available now through the GLM Coding Plan at z.ai/subscribe and works with coding agents including ZCode, Claude Code, and OpenCode. Model weights are scheduled to go open source two weeks after launch, once safety evaluation and hardening are complete.

FAQ

What is GLM-5.3?

GLM-5.3 is a coding model released by Zhipu AI on August 14, 2026. It uses the same base model as GLM-5.2, with all improvements coming from extended post-training on long-horizon task environments.

How does GLM-5.3 perform on coding benchmarks?

On the in-house Z.ai Code Bench, GLM-5.3 improves 50% over GLM-5.2. It also reaches the top open-source score on Terminal Bench 3.0 (28.3, up from 4.6) and Agents’ Last Exam (28.5, up from 23.8), and leads on DeepSWE v1.1 with 66.9.

What cybersecurity capabilities does GLM-5.3 have?

Zhipu trained GLM-5.3 with vulnerability discovery data and environments, and the model began reasoning across multiple stages of exploitation. It scored 84.5% on CyberGym and, working with security teams in China, identified 2,436 vulnerabilities across 269 projects, some dating back about 40 years.