Qwen models
All 54 Qwen models on OpenDunes, priced in Algerian Dinars.
54 models
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.
#6 for realistic voice · Qwen-Audio-3.0-TTS Flash 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 fast tier of the two Qwen-Audio-3.0 speech models, made by Alibaba. Pick it for volume work, such as many short lines, where response time matters more than the last bit of voice quality.
Qwen Image 2.1 Turbo is the fast, lower-cost tier of the Qwen Image 2.1 series, generating and editing pictures from text and image prompts where throughput and price matter more than peak quality
Qwen3 ASR 0.6B listens to audio and writes back the transcript, identifying the language itself. It handles 30 languages plus 22 Chinese dialects, on a finished file or on live speech, and returns timing for each segment and each word. Qwen describes it as the efficient trade-off, built for speed and a small footprint, so pick it for large volumes of audio.
Qwen3 ASR 1.7B listens to audio and writes back the transcript, identifying the language itself. It handles 30 languages plus 22 Chinese dialects, on a finished file or on live speech, and returns timing for each segment and each word. Qwen presents it as the more accurate of the two Qwen3 transcription models, so pick it for interviews, meetings and calls where mistakes cost you.
Qwen Image 3 is one model for making pictures from a written description and for editing them from up to four reference pictures, returning one to six pictures per request at 1K or 2K, in 13 shapes. Qwen says it draws type and detail down to 10px and knows more about the world than earlier generations. Pick it for posters, labels and interface mockups with words in them.
Qwen Image 3 Pro makes pictures from a written description, or edits them from up to four reference pictures, and returns one to six pictures per request at 1K or 2K, in 13 shapes. Like Qwen Image 3 it draws type and detail down to 10px, and it knows more about the world than earlier generations. Qwen Image 3 lists the same controls and description, so test both on your own briefs and keep Pro for the jobs where the plain version falls short.
Qwen Image 2.1 Pro makes pictures from a written description, or edits them from up to four reference pictures, and returns one to six pictures per request at 1K or 2K, in 13 shapes from tall 1:4 to wide 4:1. It is the top tier of the 2.1 series, for work where fidelity counts more than speed.
The Flash reasoning model of Qwen3.8, a multimodal all-rounder for coding help, agents, reading documents and codebases, operating a desktop, analysing charts and following long videos
An open-weight dense Qwen3.8 of 27B that sees images and video, suited to coding, professional workflows, research and long-running agent tasks, with thinking you can switch on or off
Qwen3.7 Flash reads text, pictures and video and writes back text, and it keeps about 1,000 pages in mind at once. Alibaba aims it at helpers that see and act: understanding what is in an image, writing code from a visual, searching and operating a screen, with strengths in recognizing objects and judging space.
Qwen3.6 35B A3B reads text, pictures and video and writes back text, and you can download it and run it yourself. It thinks before it answers by default, keeps that thinking across a conversation, uses the tools you connect to it and fills in the exact form or table you give it. It keeps about 260 pages in mind at once.
#3 for agent work · Qwen3.8 Max 0902 reads text, pictures and video and writes back text, and it keeps about 1,000 pages in mind at once. It is a later release of Qwen3.8 Max, trained further for coding and long tasks it works through on its own: multi-step software projects and coordinating several tools. It also reads charts, documents and long video, and it thinks before it answers by default, with an effort setting you control.
Qwen3.5 9B reads text and pictures together and writes back text, and it keeps about 260 pages in mind at once. Qwen builds it to think through problems, write code and understand images at a small size, with text and pictures handled in the same pass. It is the smallest of the Qwen3.5 models here.
Qwen3.7 Plus reads text and pictures and writes back text, and it keeps about 1,000 pages in mind at once. Alibaba calls it the cost-effective model of the Qwen3.7 series and stresses what it does with the screen: perceiving real-world scenes, reading and operating screens, writing code from a visual reference and moving through a mobile app from start to finish.
Qwen3 235B A22B Instruct 2507 reads text and writes back text and answers straight away, with no thinking phase. It keeps about 260 pages in mind at once, and Qwen singles out its understanding of many languages, its maths and its fit to open-ended requests, measured on AIME, HMMT, Arena-Hard and WritingBench.
Qwen3.8 Omni Flash reads text, pictures, sound and video together and writes back text, and it keeps about 1,000 pages in mind at once. Alibaba made it to work through tasks on its own while understanding audio and video natively, strong at long recordings that mix speech, sound and picture, and it can follow two-channel and four-channel spatial audio.
Qwen3.5 Flash reads text, pictures and video and writes back text, with fast responses and a memory of about 1,000 pages at once. It is the fast tier of Qwen3.5, a snapshot dated 23 February, and Qwen says the Flash models improve on the Qwen3 series for both text and pictures. Qwen3.6 Flash is the newer fast model.
#10 for agent work · The open-weight variant of Qwen3.8 Max, a sparse mixture-of-experts with 95B active out of 2.4 trillion, suited to coding, research, complex reasoning and agent workflows, with a million-token context
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
Qwen3 Coder Next reads text and writes back code, built for coding helpers and local development, and you can download it and run it yourself. Qwen says it matches models that do 10 to 20 times more work per word, which suits helpers that run all day, and it keeps about 250 pages in mind at once. It never shows a thinking phase and recovers from failed runs.
Qwen3 VL 235B A22B Instruct reads text and pictures and writes back text, and it is the largest model of the Qwen3 VL family here. Qwen lists answering questions about images, reading documents, pulling data out of charts and tables, reading text in many languages, operating on-screen controls and turning a sketch or mockup into code. It keeps about 260 pages in mind at once.
Qwen3 30B A3B Instruct 2507 reads text and writes back text and never shows a thinking phase, so answers come straight away. Qwen built it for following instructions, understanding many languages and working with the tools you connect to it, and it keeps about 260 pages in mind at once. Qwen reports competitive scores on reasoning, coding and writing tests such as AIME, LiveCodeBench and WritingBench.
Qwen3.8 Max Prime reads text, pictures and video and writes back text, and it keeps about 1,000 pages in mind at once. It is a version of Qwen3.8 Max built to get more work through at once, with the same abilities: thinking before it answers by default with an effort setting, using the tools you connect to it and filling in the exact form or table you give it.
Qwen3 32B reads text and writes back text, in 119 languages and dialects. A single switch makes it think before it answers, for logic, maths and code, or answer straight away in ordinary conversation. Qwen's blog credits the 14B and this 32B with general ability, coding and maths.
Qwen3.5 397B A17B reads text, pictures and video and writes back text, in 201 languages and dialects, and it thinks before it answers by default. It is the largest Qwen3.5 model in this list, which Qwen presents as covering language, reasoning, code, work done through connected tools, image and video understanding, and operating on-screen controls. It keeps about 260 pages in mind at once.
Qwen3 VL 30B A3B Instruct reads text and pictures and writes back text, with no thinking phase. Qwen lists recognizing real-world and synthetic images, locating objects in 2D and 3D, following instructions across several images, operating on-screen controls and turning sketches into working interface code, suiting document reading, text recognition and interface help.
Qwen3 Coder 30B A3B Instruct reads text, such as code and a task, and writes back code, with no thinking phase. It keeps about 250 pages in mind at once, so it can follow a whole repository, and it handles function calls, browser use and filling in code. Qwen3 Coder Next is the other model you can download and run yourself.
Qwen3.6 27B reads text, pictures and video and writes back text, in 201 languages and dialects, and you can download it and run it yourself. Qwen built it for coding helpers: understanding a whole repository, front-end work and multi-step problems, with a built-in thinking phase that carries across the conversation. It keeps about 260 pages in mind at once.
Qwen3 Coder 480B A35B reads text, such as a project's files and a task, and writes back code, using the tools you connect to it. Qwen built it for coding jobs it works through on its own across a whole repository, and it keeps about 250 pages in mind at once. Its training was 70% code and included practice on real software-fixing tasks.
The fast, efficient member of the Qwen3.6 series, reading text, images and video over a million tokens, with reasoning, tool calls and prompt caching for jobs that reuse the same context
Qwen3.7 Max reads text and writes back text, and it keeps about 1,000 pages in mind at once. It is the flagship of the Qwen3.7 series, aimed at long tasks it carries out on its own, with particular strength in coding and in office and productivity work, and it can reuse a long, repeated input without reading it again.
Qwen3.6 Plus reads text, pictures and video and writes back text, and it keeps about 1,000 pages in mind at once. Qwen reports gains over the 3.5 series in coding, front-end work and reasoning, a score of 78.8 on the SWE-bench Verified software-fixing test, and strength at 3D scenes, games and whole-repository problems.
Qwen3.5 27B reads text, pictures and video and writes back text, with fast responses. Qwen puts its overall ability at about the level of Qwen3.5 122B A10B, and it keeps about 260 pages in mind at once. It works as one whole model, where Qwen3.5 35B A3B, rated at a similar level, spreads its work across many specialist parts.
Qwen3 Next 80B A3B Instruct reads text and writes back text and answers straight away, with no thinking phase. It keeps about 260 pages in mind at once, and Qwen reports roughly 10 times the speed of Qwen3 32B on long inputs, past about 32 pages. Pick it for answering from your own long documents, for coding help and for assistants that use tools, where steady final answers matter.
A Qwen3.5 vision-language model with 10B active, second only to Qwen3.5-397B, whose text beats Qwen3-235B-2507 and whose vision beats Qwen3-VL-235B, built on linear attention with sparse experts
Qwen3 VL 8B Instruct reads text and pictures and writes back text, the smallest of the Qwen3 VL models listed here. Qwen lists document reading, answering questions about images, judging space and layout and operating on-screen controls, with text recognition in 32 languages, and it keeps about 260 pages in mind at once.
The 14B dense Qwen3, which moves between a thinking mode for maths, programming and logic and a quick mode for conversation, writing creatively and using tools in over 100 languages, up to 131K with YaRN
The 15 February snapshot of the Qwen3.5 Plus tier, a native vision-language model performing on a par with leading models in text and multimodal tasks, a large step up from Qwen3, reading images and video
The original Qwen3-30B-A3B, a mixture-of-experts with 128 experts and 3.3B active, switching between thinking for reasoning and a quick mode for dialogue, good at maths, coding and creative writing
Qwen2.5 72B Instruct reads text and writes back text, up to 8 pages per answer, in over 29 languages. Qwen lists gains in following instructions, writing long texts, reading tables and filling in the exact form or table you give it, and calls this 72B the flagship of the open Qwen2.5 line. Use it for pulling details out of documents, multilingual replies and chatbot roles set by a system message.
Qwen3 Coder Flash reads text and writes back code, working through coding jobs on its own with the tools you connect to it, and it keeps about 1,000 pages in mind at once. It is Alibaba's faster, lower-cost version of its hosted Qwen3 Coder Plus, which you cannot download, and it also handles general tasks.
Qwen2.5 7B Instruct reads text and writes back text, up to 8 pages per answer, in over 29 languages. It is the small end of the open Qwen2.5 chat line, with the same listed gains as the 72B: following instructions, reading tables and filling in the exact form or table you give it. Use it where the job is a plain instruction with a set shape of answer, such as tables in and records out.
Qwen's 72B vision-language model from the 2.5 generation, good at recognising everyday objects such as flowers, birds, fish and insects and at reading text, charts, icons and page layouts in images
Qwen-Plus reads text and writes back text, and it can use the tools you connect to it and fill in the exact form or table you give it. Alibaba builds it on Qwen2.5 and describes it as balanced between quality, speed and cost. Use it for everyday chat, summaries and forms where top quality is not needed.
Qwen3 235B A22B Thinking 2507 reads text and writes back text, and it always thinks before it answers, with room for up to about 80 pages of output. Qwen aims it at hard problems in logic, maths, science and long-form writing, citing AIME, SuperGPQA and LiveCodeBench, and it uses the tools you connect to it.
The April 2026 update of Qwen3.5 Plus, Alibaba's large multimodal model that reads text, images and video across a million tokens and writes text, with reasoning, tool calls and structured output
Qwen3 VL 30B A3B Thinking reads text and pictures and writes back text, and it thinks before it answers. Qwen presents it as the reasoning version of the family, stronger in maths, science and other complex questions about an image, with the same abilities in recognizing images, locating objects and operating on-screen controls.
The 32B Qwen2.5-Coder from the code-specific Qwen line once called CodeQwen, much better at writing, reasoning about and fixing code, and a firmer base for code agents, without losing maths and general skill
Qwen3.7 Plus reads text and pictures and writes back text, and it keeps about 1,000 pages in mind at once. Alibaba calls it the cost-effective model of the Qwen3.7 series and stresses what it does with the screen: perceiving real-world scenes, reading and operating screens, writing code from a visual reference and moving through a mobile app from start to finish.
An open-weight dense Qwen3.8 of 27B that sees images and video, suited to coding, professional workflows, research and long-running agent tasks, with thinking you can switch on or off