Qwen 3.8 Max Alibaba's flagship for long, multimodal professional work
Qwen 3.8 Max is the flagship of Alibaba's Qwen 3.8 series. It reads text, images, and video, keeps up to about a million tokens in context, and is built for coding, agent workflows, and long documents. It replaces Qwen 3.8 Max in Neurohelper AI and appears as Alibaba Qwen 3.8 Max in the Power for Professionals group of Chat Master.
Qwen 3.8 Max works inside the same Neurohelper AI workspace as Image Master, Video Master, Audio Master, and Smart Assistants. Analyze a long brief or codebase in Chat Master, then hand the plan to another model or a creative tool without switching apps.
What changed since Qwen 3.7 Max
What's new in Qwen 3.8 Max
Alibaba introduced Qwen 3.8 in early August 2026 and later published Qwen 3.8 Max in Model Studio. Compared with the Qwen 3.7 Max entry on our site, three changes matter most in day-to-day work.
It now sees images and video
Qwen 3.7 Max was a text model on our page. Model Studio lists image, text, and video input for Qwen 3.8 Max, so you can attach screenshots, charts, scans, or a short recording instead of describing them.
Longer answers
The output limit doubles to 131,072 tokens. That leaves room for complete reports, long code files, and multi-part deliverables in one reply.
Structured outputs and tools
Alibaba lists function calling, structured outputs, and web search for Qwen 3.8 Max. Structured output was not supported for the Qwen 3.7 Max alias.
Reasoning depth
How Qwen 3.8 Max reasons through hard tasks
Alibaba describes Qwen 3.8 as a 2.4-trillion-parameter mixture-of-experts model with about 95 billion parameters active per token, built on the Qwen 3.5 foundation. It thinks before it answers, and Alibaba's API offers several reasoning-effort levels, with the deepest one as the default.
Long chains of work
Plan a migration, follow a contract through every annex, or check a model against a long policy. It is designed to keep the goal and the evidence in view over many steps.
Mixed evidence
Combine a written brief with screenshots, scanned tables, or a short video and ask for one answer that cites what it saw where.
Machine-readable results
Ask for a fixed JSON schema or table when the answer feeds a spreadsheet, CRM, or script, and validate it before use.
Alibaba's API exposes reasoning-effort levels, but which level Neurohelper AI uses is set by the integration. If you need a quicker, lighter answer, switch to Qwen 3.8 Flash rather than expecting Max to skip its thinking.
How to use Qwen 3.8 Max in Neurohelper AI
Start a Qwen 3.8 Max conversation in three steps
You do not need an Alibaba Cloud account. Qwen 3.8 Max runs on the same credit balance as every other model in Neurohelper AI.
Open Chat Master
Pick Alibaba Qwen 3.8 Max from the Power for Professionals group in the model list.
Attach the real material
Paste the code or documents, and attach images or a short video where they matter. Number sources so the model can refer to them.
Ask for a checkable result
Request a format you can verify, such as a table with sources, a diff, or a JSON schema, and follow up in the same thread.
Qwen 3.7 Max → Qwen 3.8 Max
From Qwen 3.7 Max to Qwen 3.8 Max
Qwen 3.7 Max remains documented on its own page in our Archive. Here is how the two entries compare, based on Alibaba's Model Studio listings and our earlier page.
Text → text, image, and video
The 3.7 Max alias accepted text. Qwen 3.8 Max adds images and video, so visual evidence can go straight into the conversation.
65,536 → 131,072 tokens
Both keep a context window of about one million tokens, but 3.8 Max can write answers twice as long.
Structured outputs added
Alibaba lists function calling, structured outputs, web search, and context caching for Qwen 3.8 Max.
Six professional tasks
Six tasks that suit Qwen 3.8 Max
These are example prompts written for Qwen 3.8 Max, not recorded outputs. Replace bracketed parts with your own material.
Plan a refactor before you touch the code
Paste the relevant files and ask for a staged plan you can review. More engineering workflows are in our development use cases.
View the prompt
Prompt: Below are 40 files from our billing module (TypeScript, Node 22). We want to replace the legacy invoice calculator with a pure, tested pricing engine. Produce: 1) a dependency map of where the calculator is used, 2) a five-stage migration plan where every stage ships safely, 3) the test cases we must write first, and 4) risks with mitigations. Refer to files by path. Code: [paste]
Design an agent workflow with clear tool contracts
Ask for tool definitions, failure handling, and stop conditions, not just a happy path.
View the prompt
Prompt: Design an AI agent that triages incoming support emails for a SaaS company. Tools available: search_kb(query), get_customer(email), create_ticket(json), send_reply(text). Write: the system prompt, a JSON schema for each tool call, the decision flow, what the agent must never do, how it escalates to a human, and five test emails with the expected tool calls.
Compare a long set of agreements
Use the large context to keep every document in view, and ask for exact clause references so a person can check each finding.
View the prompt
Prompt: Here are our master services agreement, three order forms, and two amendments. Build a table of: payment terms, liability cap, termination rights, auto-renewal, data-processing terms, and governing law. For each, note which document controls, the clause number, and any conflict between documents. Flag anything a lawyer should review. This is not legal advice. Documents: [paste]
Turn a recording into a reproducible bug report
Attach a short screen recording and ask the model to describe exactly what happens and when.
View the prompt
Prompt: Attached is a 40-second screen recording of a checkout failing in our web app. Watch it and write a bug report with: environment visible on screen, numbered steps to reproduce with timestamps, expected vs actual result, any error text visible, and two likely causes to investigate. Mark anything you are not sure you saw.
Read a stack of screenshots for you
Attach pricing pages and landing pages and ask for a structured comparison. Pair it with our marketing use cases for next steps.
View the prompt
Prompt: Attached are 12 screenshots of pricing and landing pages from four competitors. Build a comparison matrix: plan names, prices, billing period, key limits, headline promise, main proof point, and primary CTA. Quote visible text exactly and mark anything that is cut off or unreadable. Then list three positioning gaps we could own.
Extract clean, structured data from messy inputs
Give a schema and ask for null values instead of guesses when something is missing.
View the prompt
Prompt: Extract data from the attached invoice images into this JSON schema: {vendor, invoice_number, date (ISO), currency, line_items:[{description, qty, unit_price, total}], subtotal, tax, total}. Use null when a field is missing or unreadable; do not guess. After the JSON, list any invoice where the line items do not add up to the total.
Qwen 3.8 Max vs other flagships
Qwen 3.8 Max vs DeepSeek V4 Pro vs Qwen 3.7 Max
All three are available in Neurohelper AI Chat Master or documented on our site, so you can test the same task across them. Figures come from each vendor's documentation and our DeepSeek V4 Pro and Qwen 3.8 Max pages.
| Qwen 3.8 Max | DeepSeek V4 Pro | Qwen 3.7 Max | |
|---|---|---|---|
| Developer | Alibaba | DeepSeek | Alibaba |
| Input | Text, image, video | Text | Text |
| Context window | ~1M tokens | 1M tokens | 1M tokens |
| Maximum output | 131,072 tokens | 384,000 tokens | 65,536 tokens |
| Model size (vendor) | 2.4T total / ~95B active | 1.6T total / 49B active | See Qwen 3.7 Max page |
| Best first test | Long multimodal work, structured outputs | Long text and agentic coding | Archive reference |
| In Neurohelper AI | Power for Professionals | Power for Professionals | Replaced by 3.8 Max |
Vendor figures describe each provider's own API. Context length, tools, and limits inside Neurohelper AI are set by the integration and may differ.
How to prompt Qwen 3.8 Max
Give it the evidence, the format, and the limits
A flagship reasoning model does its best work when the task is concrete. Three habits make the biggest difference.
Provide, don't describe
Paste the code, documents, or data and attach the screenshots or recording. Number every source so findings can point back to it.
Fix the output format
Name the table columns or give a JSON schema. Ask for null or "not found" instead of guesses.
State what not to do
Say which files must not change, which claims are off-limits, and when to stop and ask instead of assuming.
Use Qwen 3.8 Max when the task is long, mixed, or high-stakes. For quick drafts, summaries, and high-volume work, Qwen 3.8 Flash is the lighter choice from the same family.
Practical model selection
Where Qwen 3.8 Max is especially useful
Strong uses
- planning and reviewing changes across large codebases;
- designing agent workflows and tool-calling contracts;
- reviewing long document sets with exact references;
- analysis that mixes text with screenshots, scans, or short videos;
- extraction into strict JSON or tables;
- long reports that need more than 65K output tokens.
Review carefully when
- you need a quick answer to a simple question;
- the task is high-volume and price-sensitive;
- audio or PDF files must be read natively;
- the answer drives legal, medical, or financial decisions without review;
- you need guaranteed up-to-date facts without checking sources;
- the work requires a specific reasoning-effort setting.
Credits and plans
What Qwen 3.8 Max costs inside Neurohelper AI
Chat Master uses the same shared credit balance as image, video, and audio models. Flagship reasoning models such as Qwen 3.8 Max do more work per answer than lighter models, so long or complex requests can use more credits. Usage is shown in the app; save Qwen 3.8 Max for tasks that need it and use a lighter model for quick questions.
Start with the $1 trial
The trial costs $1 and includes 100 credits to test chat, image, video, and audio models. A card is required to start the trial.
Basic: $19 per month
Basic includes 1,000 credits a month on the monthly plan and access to every chat model in Chat Master, including the Power for Professionals group.
Pro: $49 per month
Pro includes 3,000 credits a month, more Smart Assistants, and more room for long research, document, and coding work.
Qwen 3.8 Max specifications
Compact model reference
Details are taken from Alibaba Cloud's Model Studio model list, the Qwen blog, and Alibaba's Qwen3.8-Max announcement. Context length, tools, and limits inside Neurohelper AI are set by the integration and may differ. Last reviewed November 13, 2026.
Frequently asked questions
Qwen 3.8 Max FAQ
What is Qwen 3.8 Max?
Qwen 3.8 Max is the flagship model of Alibaba's Qwen 3.8 series. It accepts text, images, and video, supports a context window of about one million tokens, and is designed for coding, agent workflows, and long professional tasks.
What is new compared with Qwen 3.7 Max?
Alibaba's Model Studio lists image and video input, a 131,072-token output limit instead of 65,536, and support for structured outputs and function calling for Qwen 3.8 Max. Qwen 3.7 Max remains documented on its own page in our Archive.
Can Qwen 3.8 Max read images and video?
Yes. Model Studio lists image, text, and video input with text output. Attach screenshots, scans, charts, or a short recording, and ask the model to say what it saw and where.
How long is the context window?
Alibaba lists a context window of about one million tokens, with up to 991,808 input tokens and up to 131,072 output tokens. Limits inside Neurohelper AI are set by the integration and may differ.
Is Qwen 3.8 Max open source?
Alibaba released the weights of its flagship Qwen 3.8 model, Qwen3.8-2.4T-A95B, on Hugging Face and ModelScope in August 2026. In Neurohelper AI you use the hosted Qwen 3.8 Max model, so you do not need your own hardware.
Qwen 3.8 Max or DeepSeek V4 Pro: which should I use?
Try Qwen 3.8 Max when the task mixes text with images or video or needs strict structured output. Try DeepSeek V4 Pro for long text and agentic coding with very long answers. Both are in the Power for Professionals group, so you can compare them on the same task.
Qwen 3.8 Max or Qwen 3.8 Flash: which should I use?
Use Max for long, complex, or high-stakes work where depth matters. Use Flash for fast drafts, summaries, and high-volume tasks where speed and cost matter more.
Can I try Qwen 3.8 Max before paying for a plan?
Yes. The $1 trial includes 100 credits and requires a card. You can use those credits in Chat Master and the other masters.
From long evidence to clear decisions
Bring the whole project to Qwen 3.8 Max.
Analyze code, documents, screenshots, and recordings in Chat Master, then turn the result into images, video, or audio in the same workspace.