Qwen: Qwen3 VL 235B A22B Thinking

What's the total on this invoice, and when is it due?

invoice-scan.jpg · photo

The total is 42 300,00 DA, due September 3, 2026. Worth a double-check: the handwritten day could read as an 8 instead of a 3.

Structured output

Turn this customer email into JSON with intent, urgency and order_id.

customer-email.txt

{ "intent": "refund", "urgency": "high", "order_id": "OD-58421" }

Full API reference Open weights on Hugging Face

Try your own request — same model, pre-selected.

Open Chat Lab

Call it in code

OpenAI-compatible — point any OpenAI SDK at our base URL and keep your code. Copy the request, or run it right here on your wallet.

EndpointPOST /v1/chat/completions

Base URL

https://opendunes.com/api/v1

Model id

qwen/qwen3-vl-235b-a22b-thinking

Shell

curl https://opendunes.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-vl-235b-a22b-thinking",
    "messages": [{"role": "user", "content": "In one sentence: why is the sky blue?"}]
  }'

The same request, in your language

Parameters you can send

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_kstopseedfrequency_penaltypresence_penaltyrepetition_penaltylogprobstop_logprobs

Reasoning

Let it think before answering — and see the thinking.

reasoninginclude_reasoning

Structured output

Get JSON you can parse instead of prose.

response_formatstructured_outputs

Tools

Let the model call your functions.

toolstool_choice

Limits

The hard numbers — what it can take in, what it can give back, and where it stops.

Context window: 131K tokens
Max output: 32.8K tokens
Longest reply: ≈ 25.0K words
Tokenizer: Qwen3
Knowledge cutoff: 2025-03-31
Released: 2025-09-23

Questions about Qwen3 VL 235B A22B Thinking

How much does Qwen3 VL 235B A22B Thinking cost in Algerian Dinars?

129,60 DA for the text you send and 1 296 DA for the text it writes back. A typical exchange — a short question and a one-screen answer — costs about 0,88 DA. You pay for what you actually use: no subscription, no minimum.

How big is Qwen3 VL 235B A22B Thinking's context window?

Qwen3 VL 235B A22B Thinking reads up to 131K tokens in one request, and writes back up to roughly 25,000 words at once.

What can Qwen3 VL 235B A22B Thinking handle?

It reads text and images and replies with text. It also supports calling your own tools, replying in structured JSON, step-by-step reasoning and streaming its answer as it writes.

Can I use Qwen3 VL 235B A22B Thinking without an international bank card?

Yes. Top up your OpenDunes balance in dinars with a CIB or Edahabia card, then spend it on Qwen3 VL 235B A22B Thinking or any other model in the catalog from that same balance. No foreign card, no currency conversion, no subscription.

Who makes Qwen3 VL 235B A22B Thinking?

Qwen3 VL 235B A22B Thinking is built by Qwen, released on Sep 23, 2025. Its training data runs to 2025-03-31. OpenDunes serves it unchanged — same model, same weights, billed in dinars.

Other Qwen models, priced in dinars

See all Qwen models

Qwen: Qwen3.8 Flash

qwen/qwen3.8-flashText

Qwen3.8 Flash is a multimodal reasoning model from Alibaba. It is suited for coding assistance, agentic workflows, visual understanding, document and codebase analysis, desktop interaction, chart analysis, and long-video analysis.

by QwenAug 26, 20261M contextReasoningToolsJSON

TextImagesVideoText

in48,60 DAout152,28 DAper 1M tokens

by QwenAug 26, 20261M contextReasoningToolsJSON

Qwen: Qwen3.8 27B

qwen/qwen3.8-27bText

Qwen3.8 27B is an open-weight dense vision-language model from Qwen. It is suited for coding, professional workflows, research, multimodal interaction, and long-running agent tasks, with flexible thinking that can be...

by QwenAug 14, 20261M contextReasoningToolsJSON

TextImagesVideoText

in137,70 DAout826,20 DAper 1M tokens

by QwenAug 14, 20261M contextReasoningToolsJSON

Qwen3 Reranker 8B

qwen/qwen3-reranker-8bRerank

Qwen3 Reranker 8B is a text reranking model from Alibaba Cloud built on the Qwen3 architecture. It evaluates query-document pairs to produce relevance scores for use in retrieval and RAG...

in64,80 DAoutFreeper 1M tokens

TextRanking

qwen/qwen3-asr-1.7bTranscription

Qwen: Qwen3 ASR 1.7B

Qwen3 ASR 1.7B is an automatic speech recognition model from Qwen. It supports multilingual language identification and transcription across 30 languages and 22 Chinese dialects, with streaming and offline inference...

2 430 DA/M input·0 DA/M output

qwen/qwen3-asr-0.6bTranscription

Qwen: Qwen3 ASR 0.6B

Qwen3 ASR 0.6B is a compact automatic speech recognition model from Qwen. It supports multilingual language identification and transcription across 30 languages and 22 Chinese dialects, with streaming and offline...

1 078,92 DA/M input·0 DA/M output

qwen/qwen3.8-2.4t-a95bText

Qwen: Qwen3.8 2.4T A95B

Qwen3.8 2.4T A95B is an open-weight sparse mixture-of-experts model from Qwen and the open-weight variant of Qwen3.8 Max, with 95 billion active parameters out of 2.4 trillion total. It is...

qwen/qwen3.8-2.4t-a95b:batchText

Qwen: Qwen3.8 2.4T A95B (batch)

Qwen3.8 2.4T A95B is an open-weight sparse mixture-of-experts model from Qwen and the open-weight variant of Qwen3.8 Max, with 95 billion active parameters out of 2.4 trillion total. It is...

Qwen: Qwen Image 3

qwen/qwen-image-3Image

Qwen Image 3 is a unified image generation and editing model from Qwen. It supports precise rendering of text and details as small as 10px, along with a richer world...

TextImagesImages

9,72 DA / image

What it can do

Real jobs this model can take on, based on what it can read and how it’s priced.

Turn a screenshot of a design into working code

understands images

Run an agent that calls your own tools and replies in clean JSON

tool calling · structured output

Work through a hard problem step by step, not just answer it

step-by-step reasoning

Power a chat product at high volume, priced per reply

≈ 0,88 DA per typical reply

Good for

Tool calling and structured JSON, for building agents
High-volume workloads where cost per reply matters

Not for

Generating images — it replies with text only
Speaking its replies aloud — no audio output
Fine-tuning on your own data — not supported

Also supports

Tool callingStructured JSONStep-by-step reasoningStreamingPrompt cachingFine-tuning

Where your request runs

You call one endpoint. Behind it, OpenDunes keeps every one of these live — if one degrades, the next takes the request without you changing a line.

Provider	In /M	Out /M	Cache /M	Latency	TPS	Uptime
Alibaba	129,60 DA	1 296 DA	—	—	—	77,9 %
Novita	317,52 DA	1 279,80 DA	—	—	—	100,0 %

Routed through the Bifrost gateway · prices in Algerian Dinars per 1M tokens · you’re charged the price of the endpoint that served your request.

What it costs

Everyday jobs, priced. Each request charges exactly what it uses — no subscription, no minimum.

Answer a quick question≈ 0,22 DA

Reply to a customer email≈ 0,56 DA

Read an invoice photo and pull out the total≈ 0,49 DA

Translate a full page of text≈ 0,67 DA

Write a one-page draft≈ 0,91 DA

Top up points

3 000 DA ≈ 3 404 typical replies · the exact charge appears after every request, down to the centime.

See it in action

Real requests, real replies. Try any of them yourself — one click, pre-filled.

Understands images

What's the total on this invoice, and when is it due?

invoice-scan.jpg · photo

The total is 42 300,00 DA, due September 3, 2026. Worth a double-check: the handwritten day could read as an 8 instead of a 3.

Structured output

Turn this customer email into JSON with intent, urgency and order_id.

customer-email.txt

{ "intent": "refund", "urgency": "high", "order_id": "OD-58421" }

Full API reference Open weights on Hugging Face

Try your own request — same model, pre-selected.

Open Chat Lab

Call it in code

OpenAI-compatible — point any OpenAI SDK at our base URL and keep your code. Copy the request, or run it right here on your wallet.

EndpointPOST /v1/chat/completions

Base URL

https://opendunes.com/api/v1

Model id

qwen/qwen3-vl-235b-a22b-thinking

Shell

curl https://opendunes.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-vl-235b-a22b-thinking",
    "messages": [{"role": "user", "content": "In one sentence: why is the sky blue?"}]
  }'

The same request, in your language

Parameters you can send

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_kstopseedfrequency_penaltypresence_penaltyrepetition_penaltylogprobstop_logprobs

Reasoning

Let it think before answering — and see the thinking.

reasoninginclude_reasoning

Structured output

Get JSON you can parse instead of prose.

response_formatstructured_outputs

Tools

Let the model call your functions.

toolstool_choice

Limits

The hard numbers — what it can take in, what it can give back, and where it stops.

Context window: 131K tokens
Max output: 32.8K tokens
Longest reply: ≈ 25.0K words
Tokenizer: Qwen3
Knowledge cutoff: 2025-03-31
Released: 2025-09-23

Questions about Qwen3 VL 235B A22B Thinking

How much does Qwen3 VL 235B A22B Thinking cost in Algerian Dinars?

How big is Qwen3 VL 235B A22B Thinking's context window?

Qwen3 VL 235B A22B Thinking reads up to 131K tokens in one request, and writes back up to roughly 25,000 words at once.

What can Qwen3 VL 235B A22B Thinking handle?

It reads text and images and replies with text. It also supports calling your own tools, replying in structured JSON, step-by-step reasoning and streaming its answer as it writes.

Can I use Qwen3 VL 235B A22B Thinking without an international bank card?

Who makes Qwen3 VL 235B A22B Thinking?

Qwen3 VL 235B A22B Thinking is built by Qwen, released on Sep 23, 2025. Its training data runs to 2025-03-31. OpenDunes serves it unchanged — same model, same weights, billed in dinars.

Other Qwen models, priced in dinars

See all Qwen models

Qwen3 Reranker 8B is a text reranking model from Alibaba Cloud built on the Qwen3 architecture. It evaluates query-document pairs to produce relevance scores for use in retrieval and RAG...

in64,80 DAoutFreeper 1M tokens

TextRanking

qwen/qwen3-asr-1.7bTranscription

Qwen: Qwen3 ASR 1.7B

2 430 DA/M input·0 DA/M output

qwen/qwen3-asr-0.6bTranscription

Qwen: Qwen3 ASR 0.6B

1 078,92 DA/M input·0 DA/M output

qwen/qwen3.8-2.4t-a95bText

Qwen: Qwen3.8 2.4T A95B

Qwen3.8 2.4T A95B is an open-weight sparse mixture-of-experts model from Qwen and the open-weight variant of Qwen3.8 Max, with 95 billion active parameters out of 2.4 trillion total. It is...

qwen/qwen3.8-2.4t-a95b:batchText

Qwen: Qwen3.8 2.4T A95B (batch)

Qwen3.8 2.4T A95B is an open-weight sparse mixture-of-experts model from Qwen and the open-weight variant of Qwen3.8 Max, with 95 billion active parameters out of 2.4 trillion total. It is...

Qwen: Qwen Image 3

qwen/qwen-image-3Image

Qwen Image 3 is a unified image generation and editing model from Qwen. It supports precise rendering of text and details as small as 10px, along with a richer world...

TextImagesImages

9,72 DA / image