GLM-5.3 · prompting-guide
Separate API, Coding Plan, benchmark conditions, and weight status instead of presenting an open-weight promise as a download.
GLM-5.3 is Z.ai's 2026-08-14 update to the GLM-5 family, aimed squarely at coding agents, tool use, automation, and defensive security work. The important change is not a new base architecture—the GLM-5.2 base is retained, with expanded post-training (longer and more realistic agent environments, harder tasks, more complete trajectories, and stronger verification). "This is a release about making a large model behave better inside real workflows, not merely making it answer isolated prompts better."
| Item | Status at launch |
|---|---|
| Release date | 2026-08-14 |
| Developer | Z.ai (Zhipu International) |
| API model ID | glm-5.3 |
| Main technical change | Expanded post-training on the GLM-5.2 base |
| Reasoning mode | Thinking required |
| Reasoning levels | low, high, max (max is the documented default) |
| Primary workloads | Coding, terminal agents, tool use, automation, research, and network defense |
| Access | GLM Coding Plan, ZCode, and documented coding-agent integrations |
| API pricing | The official pricing page had no GLM-5.3 entry at launch (do not estimate using 5.2 pricing) |
| Open weights | About two weeks after launch (pending security assessment and hardening); unavailable on launch day |
| Parameters | The GLM-5.2 base description carries over: approximately 744B total parameters / 40B active (not a new-architecture claim for 5.3) |
| Context | 5.2 documentation claims up to 1M; benchmark footnotes use different settings of 300K / 400K / 1M |
| Benchmark | GLM-5.3 | GLM-5.2 | Kimi K3 | DS V4 Pro 0813 | Fable 5 | GPT-5.6 Sol |
|---|---|---|---|---|---|---|
| Terminal-Bench 3.0 | 28.3 | 4.6 | 17.4 | n/a | 33.7 | 34.6 |
| DeepSWE 1.1 | 66.9 | 46.2 | 67.5 | 62.7 | 69.7 | 72.7 |
| NL2Repo | 58.0 | 48.9 | 58.0 | 61.1 | n/a | n/a |
| CyberGym | 84.5 | 77.2 | 80.0 | 83.3 | 83.8 | 83.6 |
| Toolathlon Verified | 73.0 | 59.9 | 76.5 | 74.1 | 74.7 | 74.9 |
| AutomationBench 1.0.6 | 48.2 | 26.2 | 46.7 | 43.2 | 46.2 | 45.8 |
| Agents' Last Exam | 28.5 | 23.8 | 27.6 | 25.7 | 23.8 | 28.6 |
| HLE with Tools | 62.5 | 54.7 | 59.8 | 60.0 | 63.9 | 64.5 |
| GDPval-AA v2 | 1769 | 1508 | 1682 | 1590 | 1743 | 1730 |
Improvements from 5.2 to 5.3: Terminal-Bench 3.0 +23.7, DeepSWE +20.7, AutomationBench +22.0, Toolathlon +13.1, CyberGym +7.3, and GDPval +261.
Limitation statement: benchmark owners, time budgets, contexts, scoring rules, and harnesses differ; a one-point difference cannot support a decisive conclusion; DeepSWE 66.9 and Kimi K3 67.5 are close and require run-level data.
API model ID: glm-5.3; thinking required; three reasoning_effort levels.
GLM Coding Plan: All tiers (Lite/Pro/Max) include GLM-5.3, with monthly fees starting at $18; credits are not standalone API token pricing.
Claude Code / OpenCode integration path: Provided in the official documentation (Claude Code can use the glm-5.3[1m] suffix + a 1M compressed context window).
Weights not yet released: Use the future tense when writing "open weights"; local deployment is currently infeasible (there is no official checkpoint or serving recipe).
Validation guidance: Teams should test the post-training narrative against their own repositories, permissions, tool schemas, and retry strategies rather than trust the launch tables alone.
"Thinking is required, with reasoning_effort values of low, high, or max; Z.ai documents max as the default."
"Do not copy GLM-5.2 pricing into a GLM-5.3 cost calculator or assume the two are identical."
"Calling GLM-5.3 'open source' on launch day is premature. … The precise description today is API-available, open-weight… This is a necessary excerpt; read the original source for full context.
Atoms.dev Blog (AI model aggregation and guide site) · Source date: 2026-08-16 · Edited: 2026-09-20
Read the original sourceGLM-5.3
Run this guide in the environment listed above. Downloading does not transfer the template or establish model availability for your account.
