GLM-5.3 API — Zhipu Flagship Quickstart

Direct answer (August 2026)

Zhipu's current flagship = GLM-5.3 (released Aug 14, 2026 — same base as GLM-5.2, all gains from post-training scaling; Terminal-Bench 3.0 4.6→28.3, CyberGym 84.5%). Z.ai GA'd the API on Aug 19 at $1.40 / $4.40 per 1M tokens ($0.26 cached) and published open weights on Aug 28 (zai-org/GLM-5.3, custom license). Target ID via ChinaModelAPI: glm-5.3 against https://api.chinamodelapi.com/v1, live once the relay health-check completes — glm-5.2 / glm-5.1 keep resolving today; confirm the live alias in your dashboard. Official product surface: z.ai / chat.z.ai. Also new: GLM-5.3-Flash — open weights at ~1/10 flagship price.

Sources: Z.ai blog “GLM-5.3” (2026-08-14); Z.ai blog “GLM-5.2: Built for Long-Horizon Tasks” (2026-06-16); zai-org/GLM-5.3 Hugging Face release (2026-08-28).

What changed vs older GLM lines

• GLM-5.3 — current flagship (Aug 14, 2026): same base as 5.2, all gains from post-training scaling; Terminal-Bench 3.0 4.6→28.3, CyberGym 84.5%; weights open since Aug 28 (custom license); GLM-5.3-Flash sibling ships open weights at ~1/10 price
• GLM-5.2 — open-weights long-horizon SWE & agents, ~1M context, MIT weights; the ChinaModelAPI production ID that keeps resolving today
• GLM-5.1 — previous 5.x step (still useful for cost/latency A/B)
• Older marketing labels like GLM-4.5 / 4-Plus are superseded for “latest” positioning — prefer the 5.x line (current flagship: GLM-5.3) in product copy

Strengths

• Long multi-step coding loops / harness-style agents (Terminal-Bench 3.0 28.3)
• Emergent cybersecurity capabilities (CyberGym 84.5%) — use responsibly
• Competitive open-weight frontier quality at lower $ than closed US APIs
• Self-host option when data must stay on-prem (custom glm-5.3 license)
• Bilingual ZH/EN product UX remains strong

Python

from openai import OpenAI
client = OpenAI(api_key="YOUR_KEY", base_url="https://api.chinamodelapi.com/v1")
resp = client.chat.completions.create(
    model="glm-5.3",
    messages=[
        {"role": "system", "content": "You are a senior software engineer. Plan then implement."},
        {"role": "user", "content": "Add retry + timeout to this HTTP client and write tests."},
    ],
    temperature=0.2,
)
print(resp.choices[0].message.content)

Node.js

import OpenAI from "openai";
const client = new OpenAI({
  apiKey: process.env.CHINAMODELAPI_KEY,
  baseURL: "https://api.chinamodelapi.com/v1",
});
const resp = await client.chat.completions.create({
  model: "glm-5.3",
  messages: [{ role: "user", content: "Refactor this service for idempotent webhooks." }],
});
console.log(resp.choices[0].message.content);

Routing tip

Default production coding/agent traffic to glm-5.3 once the relay ID completes health-check — glm-5.2 / glm-5.1 keep resolving meanwhile; confirm live aliases in your dashboard. Keep short bulk classification on cheaper IDs such as the GLM-5.3-Flash tier. For extreme multi-doc packs, also evaluate Kimi K3.

FAQ

Related Guides