Z.AI GLM 5.3: How to Access the New Flagship (Coding Plan, ZCode, API)

Quick answer: Z.AI's GLM 5.3 (released August 14, 2026) is available three ways: the GLM Coding Plan (Lite $18/mo, Pro $80/mo, Max $168/mo — all subscribers auto-upgraded), ZCode (Z.AI's coding agent), and the API (model id glm-5.3, $1.40/$4.40 per 1M, rolling out now). The same post-training-only model that jumped Terminal-Bench 3.0 from 4.6 to 28.3 and hit CyberGym 84.5 (best public result).

TL;DR

What Is Z.AI's GLM 5.3?

GLM 5.3 is Z.AI's (formerly Zhipu AI) newest flagship, released August 14, 2026. The headline: it uses the same base model as GLM 5.2 (~743B MoE per community estimates), with every gain coming from scaled post-training — more RL environments, more compute. Results:

• +50% coding on Z.AI's internal code bench
• Terminal-Bench 3.0: 28.3 (from 4.6 — open-weights SOTA)
• CyberGym: 84.5 — best public vulnerability-discovery result
• 2,436 real-world vulnerabilities found across 269 projects

Access Path 1: GLM Coding Plan (Easiest)

All GLM Coding Plan subscribers were upgraded to GLM 5.3 on launch day — no action needed.

Plan notes:

• Points-based quota — input, cached input, and output metered separately.
• Off-peak = 50% points — outside 14:00–18:00 UTC+8 weekdays (all weekends included).
• Works in ZCode, Claude Code, Cline, Kilo Code, OpenCode and 20+ tools via npx @z_ai/coding-helper.

Access Path 2: ZCode (Z.AI's Agent)

ZCode is Z.AI's native coding agent and the best showcase for GLM 5.3's long-horizon strength:

• 98%+ cache hit rate (~30% more effective tokens)
• 1.5× quota boost through August 31 (stacks to ~180% of standard quota)
• Goal mode — plans, codes, tests, verifies until done
• Remote Control — steer long tasks from WeChat/Feishu

Access Path 3: API (Rolling Out)

The glm-5.3 API is coming online now. Official pricing is $1.40 per 1M input / $4.40 per 1M output (same as GLM 5.2), confirmed on Z.AI's pricing page. Endpoints:

One required change from 5.2: thinking is always on.

{
  "model": "glm-5.3",
  "thinking": { "type": "enabled" },
  "reasoning_effort": "max"
}

reasoning_effort: low / high / max (default max; max recommended for coding). thinking.type: "disabled" will fail — migrate by setting enabled + low before switching the model ID.

Access Path 4: Open Weights (Soon)

Z.AI plans to release weights on HuggingFace ~2 weeks after launch (late August), after safety evaluation and hardening — the cyber capability is dual-use, so the release is gated. Same base as 5.2 means your self-hosting stack mostly carries over.

Which Access Path Should You Pick?

• Individual devs: Coding Plan Pro ($80/mo, or $56/mo annual) — best quota-per-dollar, priority access.
• Agent-heavy workflows: ZCode before Aug 31 (1.5× boost), then evaluate.
• Product builders: API (glm-5.3) — $1.40/$4.40 is roughly half Kimi K3's price at equal AA index score (60).
• High-volume/private: wait ~2 weeks for weights and self-host.

FAQ

What is Z.AI GLM 5.3? Z.AI's newest flagship (August 14, 2026) — same base as GLM 5.2, all gains from post-training: +50% coding, Terminal-Bench 3.0 SOTA, emergent cyber capability.

How do I access GLM 5.3? GLM Coding Plan (auto-upgraded), ZCode, or the API (glm-5.3, rolling out). Weights in ~2 weeks.

How much does GLM 5.3 cost? API: $1.40/$4.40 per 1M tokens. Coding Plan: $18/$80/$168 per month (annual −30%).

Is GLM 5.3 open weights? Planned — HuggingFace ~2 weeks after launch, pending safety evaluation.

What changed in the API? Thinking is always enabled; reasoning_effort (low/high/max) replaces thinking.type: "disabled".

Sources

• Z.AI: GLM-5.3: Frontier Coding with Emergent Cyber Capabilities (August 14, 2026)
• Z.AI Docs: GLM-5.3
• Z.AI Docs: Pricing

Last updated: August 14, 2026