Need a China coding plan (GLM / MiniMax / Kimi / Volcengine / Xiaomi)? Email me at [email protected]
Zhipu AI · Z.ai
GLM-5.3 (Terminal-Bench 3.0 #1 open-source) and GLM-5.3-Flash (native multimodal), both with a genuinely usable 1M context, are available on all tiers. Credits refresh on a 5-hour + weekly dual window, with 50% off-peak deduction; quarterly 20% and yearly 30% discounts. ZCode is now a standing benefit (150% plan quota, free off-peak tasks) and AutoClaw joins as the official office assistant.
All plans support GLM-5.3 and GLM-5.3-Flash, both with a genuinely usable 1M context; calls to legacy models are auto-routed to the new versions.
Newest flagship, live for all users since Aug 14. Same base as 5.2 with post-training scaling: +50% coding on Z.ai Code Bench, #1 open-source on Terminal-Bench 3.0 (28.3), DeepSWE v1.1 at 66.9. Weights released on HuggingFace Aug 28, and it supports a genuinely usable 1M context.
Live Aug 26: the first natively multimodal model in the GLM-5 family (320B total / 18B active params, hybrid linear+sparse attention). DeepSWE v1.1 at 63.4 beats GLM-5.2; coding/agent near Claude Opus 4.8. Consumes 3x the usage of GLM-5.3 (lower Credits rate). Weights open-sourced under MIT, and it also supports a genuinely usable 1M context.
Calls to GLM-5.2 / GLM-5.1 automatically switch to GLM-5.3; GLM-5-Turbo / GLM-4.7 switch to GLM-5.3-Flash — no config change needed.
Monthly prices with quarterly 20% / yearly 30% off. Credits refresh on a 5-hour + weekly dual window; off-peak usage (outside Mon-Fri 14:00-18:00 UTC+8) deducts 50%. ZCode 150% quota and AutoClaw are subscriber benefits on every tier.
Lite
Yearly $12.6/mo (-30%)
Individual developers and small repos
Pro
Best valueYearly $56/mo (-30%)
Daily developers working on medium repos
Max
Yearly $117.6/mo (-30%)
Heavy users and teams needing guaranteed capacity
| Feature | Lite | Pro | Max |
|---|---|---|---|
| Monthly price | $18 | $80 | $168 |
| Yearly (-30%) | $12.6/mo | $56/mo | $117.6/mo |
| Credits / 5h | 2,000 | 12,000 | 28,000 |
| Credits / week | 10,000 | 60,000 (6x) | 140,000 (14x) |
| GLM-5.3 / 5.3-Flash | Yes | Yes | Yes |
| Off-peak deduction | 50% off | 50% off | 50% off |
Peak hours are Mon-Fri 14:00-18:00 (UTC+8); everything else deducts 50% fewer Credits — Z.ai claims up to 92% savings vs pay-as-you-go GLM-5.3 API.
Visual understanding, web search, web reading and open-source repo access (1.2 Credits per call).
Weekends and weekday nights deduct half Credits — schedule heavy jobs accordingly.
Night unlimited (Sep 3 - Oct 7, 23:00-09:00): GLM-5.3-Flash calls from ZCode / AutoClaw are uncapped and other agents get double quota. Double Festival (Sep 25 - Oct 7): off-peak deduction applies all day.
Standing subscriber benefits: ZCode gives 150% plan quota plus free off-peak tasks and professional data MCP; AutoClaw, the new official office assistant, gets its own 150% quota and dedicated perks.
Invite program returns 10-20% Credits (overseas).
GLM-5.3-Flash weights are open under MIT; GLM-5.3 flagship weights landed on HuggingFace Aug 28.
Lite $18/mo (yearly $12.6/mo), Pro $80/mo (yearly $56/mo), Max $168/mo (yearly $117.6/mo). Quarterly billing gives 20% off, yearly 30% off.
Credits refresh on a dual window: 5-hour (Lite 2,000 / Pro 12,000 / Max 28,000) plus weekly (10,000 / 60,000 / 140,000). Off-peak usage (outside Mon-Fri 14:00-18:00 UTC+8) deducts 50% fewer Credits; flagship models consume Credits faster.
A new model live Aug 26: the first natively multimodal model in the GLM-5 family (320B total / 18B active parameters). DeepSWE v1.1 at 63.4 beats GLM-5.2 and coding/agent capability approaches Claude Opus 4.8. It uses 3x the usage of GLM-5.3 (a lower Credits rate) and its weights are open-sourced under MIT.
All tiers support GLM-5.3 (Terminal-Bench 3.0 #1 open-source) and GLM-5.3-Flash (native multimodal), both with a genuinely usable 1M context. Calls to legacy GLM-5.2/5.1 auto-route to GLM-5.3, and GLM-5-Turbo/4.7 auto-route to GLM-5.3-Flash.
ZCode is a standing benefit rather than a temporary bump: subscribers get 150% plan quota, free off-peak tasks and a professional data MCP, and AutoClaw (the official office assistant) gets its own 150% quota. Two limited-time events run through early October: night unlimited GLM-5.3-Flash via ZCode / AutoClaw plus double quota for other agents (Sep 3 - Oct 7, 23:00-09:00), and the Double Festival, when the off-peak rule applies all day (Sep 25 - Oct 7). Regular off-peak deduction stays at 50% outside Mon-Fri 14:00-18:00 UTC+8.
Yes. GLM-5.3-Flash weights are open under MIT, and the GLM-5.3 flagship weights were released on HuggingFace on Aug 28 as promised (two weeks after launch).
No. Visual understanding, web search, web reading and open-source repo MCP tools are included in every tier.