by QwenReleased 25 February 2026
The 35B-A3B Qwen3.5, a native vision-language model with only 3B active that uses linear attention and sparse experts for efficient inference, overall on a par with Qwen3.5-27B
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 | 46,60 DA | 332,84 DA |
| Cached tokens read | 16,64 DA |
per 1M tokens
Seven companies run this model. OpenDunes sends each request to one of them, picked by price and speed. If it fails, the next one takes it. DA per 1M tokens: ↑ what you send · ↓ what comes back.
| Provider | ↑ Sent | ↓ Back | Cache read | Latency | Throughput | Uptime |
|---|---|---|---|---|---|---|
| Darkbloom | 21,63 DA | 249,63 DA | 10,82 DA | … | … | … |
| DeepInfraUnited States | 46,60 DA | 332,84 DA | 16,64 DA | … | … | … |
| Parasail | 49,93 DA | 332,84 DA | 16,64 DA | … | … | … |
| VeniceUnited States | 49,93 DA | 332,84 DA | 16,64 DA | … | … | … |
| AlibabaAsia | 54,09 DA | 432,69 DA | none | … | … | … |
| AtlasCloud | 74,89 DA | 599,10 DA | 74,89 DA | … | … | … |
| SiliconFlowUnited States | 79,88 DA | 599,10 DA | 49,93 DA | … | … | … |
You pay the price of the provider that served your request. Prices read 15:20 · speed over the last 30 minutes · uptime over 3 days.
Real requests, real replies. Try any of them yourself — one click, pre-filled.
What's the total on this invoice, and when is it due?
invoice-scan.jpg · photoThe total is 42 300,00 DA, due 5 September 2026. Worth a double-check: the handwritten day could read as 8 instead of 3.
Try in Chat LabTurn this customer email into JSON with intent, urgency and order_id.
customer-email.txtTry your own request — same model, pre-selected.
Open Chat LabSwap 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="qwen/qwen3.5-35b-a3b",
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_biaslogprobstop_logprobsReasoning
Let it think before answering — and see the thinking.
reasoninginclude_reasoningStructured output
Get JSON you can parse instead of prose.
response_formatstructured_outputsTools
Let the model call your functions.
toolstool_choiceAbout 0,21 DA per message. ↑ 46,60 DA · ↓ 332,84 DA per 1M tokens.
No. You pay per use from your balance, with no subscription. Try it in Chat at the same price.
The 35B-A3B Qwen3.5, a native vision-language model with only 3B active that uses linear attention and sparse experts for efficient inference, overall on a par with Qwen3.5-27B.
Deposit dinars with your Edahabia or CIB card. Every model, Qwen3.5-35B-A3B included, is paid from that balance. No foreign card, no conversion.
Qwen states support for 201 languages and dialects but does not list them in the model card. Arabic, French and Darja are not named.
It reads text, images and video. Up to 262K tokens in one request, about 260 pages. It writes up to 61 000 words at once.
Yes, its weights are public. Open weights on Hugging Face
Qwen. Released 25 February 2026.
The 8B Qwen3 Embedding, carrying Qwen3's multilingual reach and long-text understanding into vectors for text and code retrieval, classification, clustering and bitext mining; the 4B is its smaller sibling
Qwen3 Reranker 8B reads a search question together with each candidate text and gives each pair a relevance score, so the best matches rise to the top. It covers over 100 languages and programming languages, and you can tell it what counts as relevant for your task. Use it as the second pass after a search step such as Qwen3's text-meaning models has fetched candidates, to put your answers in order. It does not write text itself.
Qwen3 Embedding 4B turns text into a fingerprint of its meaning, so search finds text that means the same even in other words or another language. It covers over 100 languages, code included, and reads about 32 pages at once, which suits searching documents or code, grouping similar texts and matching translations.
Qwen-Audio-3.0-TTS Plus reads text and speaks it aloud, and Alibaba's docs say it takes plain-language directions for speech rate, emotion and speaking style. It is the higher-quality tier of the two Qwen-Audio-3.0 speech models, made by Alibaba. Pick it for narration, voiceover or any line where voice quality matters more than speed.
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.
Read a long report with charts and answer questions
Images
Write and fix code with your test tool in the loop
Uses your tools
Summarise a short product video
Video