GLM-5.3 API Is Live (August 2026): Pricing, Access & What Changed vs GLM-5.2

Updated August 25, 2026. Availability verified with a live API call.

GLM-5.3 shipped on August 14, and early access was initially limited to Coding Plan subscribers. That changed within days: the open API is now live for everyone, and Zhipu priced it exactly the same as GLM-5.2. If you've been waiting to move off 5.2, there is no pricing reason to wait any longer.

Current status: the open API is live

We verified this with a direct call on August 25. The glm-5.3 model ID appears in the standard /v4/models listing, and chat/completions returns HTTP 200 with a normal completion. One thing worth knowing before you integrate: GLM-5.3 is a reasoning model. Responses carry a reasoning_content block alongside the visible answer, and reasoning tokens are billed as output. Give it a comfortable token budget (we recommend max_tokens >= 400) or short replies can get truncated mid-thought.

Pricing: same as GLM-5.2, cheaper than OpenRouter

Official list price: ¥8 / 1M input tokens, ¥28 / 1M output tokens (cache hit ¥2) — identical to GLM-5.2.
OpenRouter: $1.40 / $4.40 per 1M tokens.
NovAI: $1.25 / $4.00 per 1M tokens — about 11% below OpenRouter, with zero platform fee.

There is no glm-5.3-air variant at launch; calling that model ID returns a 400. The standard glm-5.3 endpoint is the only way in.

What changed vs GLM-5.2

1M-token context window (up from 200K on 5.2) with up to 128K-token outputs.
Always-on reasoning: every response includes a chain-of-thought trace, which helps on multi-step coding and agentic workloads but adds latency and output tokens.
Stronger coding and agentic performance, pitched by Zhipu at long-horizon autonomous tasks and security-oriented work.
Same price — the upgrade is free if you were already paying 5.2 rates.

How to call it

import requests

resp = requests.post(
    "https://aiapi-pro.com/v1/chat/completions",
    headers={"Authorization": "Bearer YOUR_NOVAI_KEY"},
    json={
        "model": "glm-5.3",
        "max_tokens": 600,
        "messages": [{"role": "user", "content": "Refactor this function for clarity..."}],
    },
)
msg = resp.json()["choices"][0]["message"]
print(msg.get("reasoning_content", "")[:200])  # reasoning trace
print(msg["content"])

The response is OpenAI-compatible, so any SDK pointed at https://aiapi-pro.com/v1 works with a one-line model change.

Should you switch from GLM-5.2?

If your workload tolerates reasoning latency, yes — it is a strict capability upgrade at the same price. If you need fast, cheap, non-reasoning completions (classification, extraction, high-volume chat), glm-5.2 remains the better fit, and GLM-4.7-Flash covers the near-free tier.

Call GLM-5.3 right now on NovAI

$1.25 / $4.00 per 1M tokens · 1M context · OpenAI-compatible · free credit, no card required

Get your API key free →

Or test it in the playground →

NovAI proxies the official Zhipu endpoint with unified OpenAI-compatible billing — one key for GLM-5.3, GLM-5.2, Kimi, DeepSeek and 30+ more models.