by Z.AIReleased 18 August 2026
#2 for slide decksGLM-5.3 reads and writes text, keeps about 1,000 pages in mind at once, and always thinks before it answers, though you choose low, high or maximum effort, with maximum the default. Z.ai says it beats the earlier GLM-5.2 at coding and gets good results without wasting effort. Pick it for demanding programming and long jobs it carries out by itself. GLM-5.3 Prime is a faster version of it, and GLM-5.3 Flash suits work with pictures and video.
You pay for what each request uses. No subscription, no minimum.
38 % offDeepInfra has cut its price. Every figure here is the new one.
One message, the ordinary use
Every rate
| Meter | ↑ you send | ↓ comes back |
|---|---|---|
| Text tokens | 149,78 DA | 665,67 DA |
| Cached tokens read | 33,28 DA |
per 1M tokens
One company runs this model. OpenDunes sends every request to it. DA per 1M tokens: ↑ what you send · ↓ what comes back.
You pay the price of the provider that served your request. Prices read 15:19 · speed over the last 30 minutes · uptime over 3 days.
Swap the base URL and the key. Your OpenAI code runs unchanged.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://opendunes.com/v1",
api_key=os.environ["OPENDUNES_API_KEY"],
)
resp = client.chat.completions.create(
model="z-ai/glm-5.3:batch",
messages=[{"role": "user", "content": "Two lines to welcome a shop in Oran."}],
)
print(resp.choices[0].message.content)Runs this request once on your balance.
Standard OpenAI fields, grouped by what they control. Anything not listed is accepted and ignored.
Shaping the reply
How long, how varied, where to stop.
max_tokenstemperaturetop_ptop_kmin_pstopseedfrequency_penaltypresence_penaltyrepetition_penaltylogit_biasReasoning
Let it think before answering — and see the thinking.
reasoningreasoning_effortinclude_reasoningStructured output
Get JSON you can parse instead of prose.
response_formatstructured_outputsTools
Let the model call your functions.
toolstool_choiceAbout 0,48 DA per message. Was 0,77 DA. ↑ 149,78 DA · ↓ 665,67 DA per 1M tokens.
No. You pay per use from your balance, with no subscription. Run it once from Call it in code, at the same price.
#2 for slide decks · GLM-5.3 reads and writes text, keeps about 1,000 pages in mind at once, and always thinks before it answers, though you choose low, high or maximum effort, with maximum the default. Z.ai says it beats the earlier GLM-5.2 at coding and gets good results without wasting effort. Pick it for demanding programming and long jobs it carries out by itself. GLM-5.3 Prime is a faster version of it, and GLM-5.3 Flash suits work with pictures and video.
Deposit dinars with your Edahabia or CIB card. Every model, GLM 5.3 (batch) included, is paid from that balance. No foreign card, no conversion.
Z.ai lists English and Chinese. Arabic, French and Darja are not named.
It reads text. Up to 1M tokens in one request, about 1 000 pages. It writes up to 98 000 words at once.
Yes, its weights are public. Open weights on Hugging Face
Z.AI. Released 18 August 2026.
#8 for agent work · GLM-5.3 Flash takes text, images and video, writes back text, and keeps about 1,000 pages in mind at once. Z.ai built it for light, quick coding and long tasks it carries out by itself, and says it stays accurate over very long material while using less computing. Pick it when the job includes screenshots or clips and you want to keep running costs down.
#2 for slide decks · GLM-5.3 reads and writes text, keeps about 1,000 pages in mind at once, and always thinks before it answers, though you choose low, high or maximum effort, with maximum the default. Z.ai says it beats the earlier GLM-5.2 at coding and gets good results without wasting effort. Pick it for demanding programming and long jobs it carries out by itself. GLM-5.3 Prime is a faster version of it, and GLM-5.3 Flash suits work with pictures and video.
#10 for games · GLM 5.2 thinks before it answers. It reads and writes text, keeps about 1,000 pages in mind at once, and can write an answer several hundred pages long. Z.ai made it for software projects from start to finish: it keeps track of the requirements, follows the same standards throughout, and carries a job from the first brief to release on several platforms, and you can ask it to think hard or at maximum.
The high-speed FlashX edition of GLM-5.3-Flash, reaching up to 200 tokens a second on the same 320B hybrid-attention model with 18B active, for efficient coding, visual understanding and long agent tasks
From what its maker and two benchmark sites publish, and where it stands on OpenDunes. Read 7 Nov 2026.
Thinks first
Works through the problem before it answers.
Uses your tools
Can call functions your code provides.
Fills your JSON
Answers in the JSON shape you give it.
Searches the web No.
Can look things up online while it answers.
Cheaper on repeats
A prompt it saw a moment ago costs less the second time.
Open weights
The model's weights are public.
Carry a large refactor across a repository
Thinks first
Run an engineering agent with your tools
Uses your tools
Review a pull request for design problems