ChinaModelAPI

News / Model Watch · API General Availability

Official API GA GLM-5.3 Pricing unchanged Weights out ✓
2026-08-22 · Model Watch · API opened Aug 19 (Beijing)

GLM-5.3 API Goes GA: $1.40/$4.40, One-Day Slip, and an Ecosystem That Didn't Wait

Five days after GLM-5.3's Aug 14 launch (Coding Plan first, API "coming"), Zhipu opened general API access in the early hours of August 19 — one day later than planned, at prices identical to GLM-5.2. This brief has the verified timeline, the pricing table against its Chinese-model peers, the platform roll call, and what changes for API builders before the open weights land around Aug 28.

Direct answer

The GLM-5.3 API is generally available as of August 19, 2026 (Beijing, early hours), per Zhipu's official announcement carried by 证券时报 and 第一财经. Pricing is unchanged from GLM-5.2 — $1.40 / $4.40 per 1M tokens (input / output) on the docs.z.ai pricing page. It enters at Artificial Analysis Index 60, tied with Kimi K3 as open-source co-#1; open weights delivered Aug 28 — see our open-weights brief.

The launch week, verified

WhenWhat happened · source
Aug 14GLM-5.3 launches — Coding Plan / ZCode first, API promised. z.ai blog; our launch brief. PhanRouter and JD Cloud MaaS adapt same-day (量子位).
Aug 18The API doesn't open on schedule — customers who'd budgeted for launch weekend push hard (华尔街见闻: "delay一天我们的管理成本就多好几万").
Aug 19 (凌晨)API GA. Zhipu's WeChat announcement: 即日起正式上线, 定价与 5.2 持平, 权重下周五开源 (IT之家, 界面新闻). English side: VentureBeat notes the $1.4/$4.4 hold.
Aug 21China's National Supercomputing Internet platform adds GLM-5.3 API one-click calls (OpenAI- and Anthropic-compatible); Zhipu (2513.HK) jumps 11% alongside MiniMax in the "HK large-model duo" rally (ZAKER).

Pricing: unchanged, and where that lands it

Zhipu held the line at $1.40 / $4.40 per 1M tokens (cached input ~$0.26) — the same slot GLM-5.2 occupied, and still among the cheapest frontier-grade API slots available.

Model (2026)Input $/1MOutput $/1MNote
GLM-5.3 (GA)$1.40$4.40cached ~$0.26; AA 60, open co-#1
qwen3.8-max$2.00$6.00open weights since Aug 12–17
Kimi K3$3.00$15.002.8T open weights, modified license
deepseek-v4-flash (off/peak)$0.22 / $0.44$0.66 / $1.32new peak/off-peak windows since Aug 17

Comparison rows reuse prices verified in our earlier briefs; GLM row per docs.z.ai pricing page.

What shipped — and the one breaking change to remember

  • General API access. OpenAI-compatible calls per docs.z.ai (the official quickstart covers curl/Python/Node); the model is positioned for complex coding, defensive cybersecurity, and long-horizon tasks.
  • Day-one surfaces. ZCode and the GLM Coding Plan are the reference integrations; PhanRouter (same-day Aug 14), JD Cloud MaaS (Aug 14), and the National Supercomputing Internet (Aug 21, OpenAI + Anthropic compatible) are already serving it.
  • The breaking change. thinking.type: 'disabled' was removed — if your 5.2-era code sets it, drop it before switching (flagged in our launch brief).
  • Independent scores are in. AA Intelligence Index 60 — same tier as Kimi K3 for open models, near Claude Fable 5 / GPT-5.6 Sol; the earlier "very verbose" flag (170M vs 72M median tokens on AA's index run) is worth remembering for cost-sensitive agent loops.

Get GLM through one OpenAI-compatible endpoint

ChinaModelAPI is an independent relay bringing Chinese frontier models — GLM, DeepSeek, Qwen, Kimi — to builders worldwide behind a single OpenAI-compatible API with USDT/USD1 payments. On the relay, glm-5.2 / glm-5.1 remain the production IDs today; the glm-5.3 ID follows once GA is confirmed healthy through relays.

ChinaModelAPI is an independent relay with no official relationship to Zhipu / Z.ai.

Primary sources

FAQ (2026)

When did the API go live?

Aug 19, 2026, early hours Beijing time — one day after the originally expected Aug 18, per Zhipu's official announcement.

Pricing?

Same as GLM-5.2: $1.40 in / $4.40 out per 1M on docs.z.ai, cached input much lower. No GA price bump.

Open weights when?

Delivered Aug 28 (night): 141 shards + BF16 live on zai-org, custom glm-5.3 license. Our open-weights brief has the details.

Where can I call it?

Z.ai/bigmodel API, ZCode + Coding Plan, PhanRouter, JD Cloud MaaS, and the National Supercomputing Internet platform (OpenAI/Anthropic-compatible).

vs GLM-5.2 for developers?

Same base, post-training gains (Terminal-Bench 3.0 4.6→28.3; AA 60 open co-#1). Remember: thinking.type 'disabled' was removed — a breaking change.

Through ChinaModelAPI?

Not yet at publication — glm-5.2/glm-5.1 are the relay's production IDs; glm-5.3 routing follows once GA is healthy through relays. Waitlist gets first notice.

Related guides