First-batch capacity is limited. We onboard developers in the order they join the waitlist — reserve your spot now.
Limited first-batch capacity · Reserve your spot

GLM-5.2, at half the price.

Production-grade GLM-5.2 inference for developers — an OpenAI-compatible API priced at 50% of the Zhipu US list rate. Same model, lower bill.

GLM-5.2 · per 1M tokens−50%
Input$1.40 $0.70
Output$4.40 $2.20
1M in + 1M out$5.80 $2.90
Same GLM-5.2 model. Half the Zhipu US price.
$0.70 / 1M in  ·  $2.20 / 1M out100% OpenAI-compatible APISelf-hosted · low-latency SE Asia region
Why Halfstack

The same frontier model, without the frontier bill.

We host GLM-5.2 ourselves on dedicated hardware, so we pass the savings straight to you.

💸

Exactly half the price

Every token and every plan is priced at 50% of the equivalent Zhipu US list rate. No usage tiers to game, no fine print.

🔌

Drop-in compatible

OpenAI-compatible endpoints. Change the base URL and API key — your existing SDKs, agents, and tools just work.

Low-latency region

Dedicated GPU capacity hosted in Southeast Asia (Malaysia), tuned for steady throughput on long-running agent workloads.

🧱

Built for production

Streaming, function calling, and long context — the full GLM-5.2 capability set, ready for real applications.

📊

Transparent usage

A clean dashboard for keys, spend, and rate limits. Know exactly what you're using and what it costs.

🌍

For US & EU teams

Billing in USD/EUR, English support, and terms written for developers shipping to Western markets.

Pricing

One simple promise: half of Zhipu US.

Pay per token, or subscribe monthly — always exactly 50% of the equivalent Zhipu US (z.ai) list price.

Zhipu US (z.ai) list price
GLM-5.2 API
Input / 1M tokens$1.40
Cached input / 1M$0.26
Output / 1M tokens$4.40
You save 50%
Halfstack
GLM-5.2 API
Input / 1M tokens$0.70
Cached input / 1M$0.13
Output / 1M tokens$2.20
Coding Plan

A flat monthly fee for your AI coding tools — at half the price.

Use GLM-5.2 inside Claude Code, Cline, Roo Code, Kilo Code and other OpenAI-compatible agents. One monthly fee, a generous prompt quota — no per-token metering.

Coding Lite
$9/mo
Zhipu US: $18/mo
~80 prompts / 5 hrs · ~400 / week. Great for side projects and evenings in your IDE.
Coding Max
$80/mo
Zhipu US: $160/mo
~1,600 prompts / 5 hrs · ~8,000 / week. For heavy, always-on agent workloads.

Prices reflect Zhipu's published US (z.ai) list rates as of July 2026, halved. GLM-5.2 API list price is $1.40 input / $4.40 output per 1M tokens; the Coding Plan tiers are $18 / $72 / $160 per month. We peg to Zhipu's list price and charge exactly 50%. Join the waitlist and we'll send the confirmed, final price list before you're ever asked to pay.

Developer experience

If you've used the OpenAI SDK, you're already done.

chat.py
from openai import OpenAI

client = OpenAI(
    base_url="https://api.halfstack.dev/v1",
    api_key="sk-your-key",
)

resp = client.chat.completions.create(
    model="glm-5.2",
    messages=[
        {"role": "user",
         "content": "Ship it."},
    ],
    stream=True,
)
# same SDK. same code. half the cost.
  • No rewrite. Point your existing OpenAI client at our base URL.
  • Streaming & tools. Function calling and server-sent streaming out of the box.
  • Works with your stack. LangChain, LlamaIndex, Vercel AI SDK, Cline, and more.
  • Instant keys. Generate a key, set two env vars, and you're live.
FAQ

Questions, answered.

Is this really the same GLM-5.2?+

Yes. We run the GLM-5.2 model on our own dedicated GPU infrastructure. You get the full model — the difference is our cost structure, not a smaller or distilled version.

How can you charge half of Zhipu's US price?+

We self-host on dedicated hardware in Southeast Asia and run lean. That lets us peg every rate to Zhipu's published US list price and cut it in half.

API pricing or the Coding Plan — which do I want?+

Use the pay-per-token API for production apps and scripts — you pay only for what you use, with no tool restrictions. Use the Coding Plan if you live inside a coding agent (Claude Code, Cline, Roo Code…) all day — a flat monthly fee caps your bill. Both are exactly 50% of Zhipu's US price.

When does it launch?+

We're finalizing deployment now. Join the waitlist and you'll be among the first emailed when API keys open up — waitlist members get first access.

Where are the servers, and what about latency?+

Our inference runs in Malaysia on dedicated capacity. It's well-suited to agent and batch workloads; we'll publish real latency benchmarks for US/EU regions before launch.

What do I need to do to switch?+

If you already use the OpenAI SDK (or any OpenAI-compatible client), you only change the base URL and API key. No code rewrite.

Do you train on my data?+

No. Your prompts and completions are not used for training. Full data-handling terms will be published before launch.

Reserve your access

Get GLM-5.2 for half — before everyone else.

Capacity is limited at launch, and we onboard in waitlist order. The earlier you join, the sooner you get keys and the confirmed price list.