by OpenAI·Released 5 August 2025
gpt-oss-20b reads text and writes text, keeping about 130 pages in mind at once. It is the smaller of OpenAI's two downloadable models, free to run yourself under an open license, quick enough to answer fast and light enough for an ordinary computer or a single graphics card. You choose how long it thinks, it can use the tools and apps you connect, and it can be retrained on your own examples.
You pay for what each request uses. No subscription, no minimum.
One message, the ordinary use
Every rate
| Meter | ↑ you send | ↓ comes back |
|---|---|---|
| Text tokens | 7,99 DA | 37,28 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.
| Provider | ↑ Sent | ↓ Back | Latency | Throughput | Uptime |
|---|---|---|---|---|---|
| DeepInfraUnited States | 7,99 DA | 37,28 DA | … | … | … |
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="openai/gpt-oss-20b: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 0,027 DA per message. ↑ 7,99 DA · ↓ 37,28 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.
gpt-oss-20b reads text and writes text, keeping about 130 pages in mind at once. It is the smaller of OpenAI's two downloadable models, free to run yourself under an open license, quick enough to answer fast and light enough for an ordinary computer or a single graphics card. You choose how long it thinks, it can use the tools and apps you connect, and it can be retrained on your own examples.
Deposit dinars with your Edahabia or CIB card. Every model, gpt-oss-20b (batch) included, is paid from that balance. No foreign card, no conversion.
OpenAI names no languages for gpt-oss-20b (batch). Arabic, French and Darja: not stated.
It reads text. Up to 131K tokens in one request, about 130 pages. It writes up to 88 000 words at once.
Yes, its weights are public. Open weights on Hugging Face
OpenAI. Released 5 August 2025. It knows the world up to 30 June 2024.
Text Embedding 3 Small turns text of up to about 8 pages into a fingerprint of its meaning, so search finds passages that mean the same thing even when the words differ. It replaces the older ada-002 and is the sensible default for search over your documents, grouping similar messages or labelling them.
GPT Image 2.5 Sunburst takes a written description, with optional reference pictures, and gives back an image, new or edited, as PNG, JPEG or WebP, with transparent backgrounds on PNG and WebP. It is the precision tier of the 2.5 series, for detailed creative work where an edit has to land exactly as asked, and it adds top quality settings named xhigh and max.
GPT-6 Luna Decisions does not write text. You send content as text, a table of data or a picture, plus up to 200 named questions of three kinds: yes or no, pick among options, or grade against an ordered scale. It returns a probability for every answer, with the likeliest option or expected score, so your own system can set its thresholds for routing, moderation or marking.
Text Embedding 3 Large turns text of up to about 8 pages into a fingerprint of its meaning, so search finds passages that mean the same thing even when the words differ. It is the most accurate of OpenAI's models of this kind, in English and other languages. Choose it when the quality of matches matters most, for search, grouping or labelling documents.
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 No.
A prompt it saw a moment ago costs less the second time.
Open weights
The model's weights are public.
Knows events up to 30 Jun 2024
What it learned stops at this date.
Run a small agent that looks up orders through your API
Uses your tools
Pull fields from emails into JSON
Fills your JSON
Work through a short logic problem step by step
Thinks first