Last verified: July 26, 2026
Independent and unofficial: Qwen AI Chat is not affiliated with, endorsed by or operated by Alibaba Cloud, QwenCloud or the Qwen team. This page verifies public information against official sources. Any chat demo on this website may use a different model or provider and is not evidence that Qwen3.8 Max Preview is running here.
Qwen3.8 Max Preview, identified by the exact model ID qwen3.8-max-preview, is a hosted Qwen AI preview that became available on July 19, 2026. It is not Qwen3-8B: “Qwen3.8” is the newer version name, while Qwen3-8B is a separate, earlier 8-billion-parameter checkpoint. Official materials currently confirm Qwen3.8 Max Preview through QwenCloud Token Plan, Qoder, QoderWork and supported developer tools. They do not establish it as a finished production release, the default model in Qwen Studio or a Fireworks AI model.
Qwen and Qoder report a total scale of 2.4 trillion parameters. QwenCloud documents reasoning, visual understanding, text generation, a 1M context category, Function Calling and built-in tools. The architecture, activated-parameter count, training data, knowledge cutoff, final license and production release date have not been published in a complete public model card.
Current status: The model is real and accessible, but it remains an evolving preview. Qwen says an open-weight Qwen3.8 release is planned; no official weights, repository, license or release date had been published when this page was checked.
Qwen3.8 Max Preview at a glance
| Item | Verified information |
|---|---|
| Official display name | Qwen3.8 Max Preview |
| Exact model ID | qwen3.8-max-preview (case-sensitive) |
| Current lifecycle | Preview; it may be updated, taken offline or replaced by a production model |
| First documented availability | July 19, 2026 |
| Parameter count | 2.4 trillion total parameters, according to Qwen/Qoder |
| Confirmed capabilities | Reasoning, visual understanding and text generation |
| Context | 1M in QwenCloud’s model table; an official OpenCode configuration uses 983,616 context tokens |
| Published output setting | 131,072 maximum output tokens in QwenCloud’s OpenCode configuration; do not assume provider parity |
| Thinking | Supported; QwenCloud’s documented Token Plan client configurations enable it |
| Function Calling | Listed as supported by QwenCloud |
| Built-in tools | Listed as supported through Token Plan Harness services |
| Structured output | Not listed as supported in QwenCloud’s current model table |
| Confirmed access | QwenCloud Token Plan, Qoder, QoderWork and Qwen Code’s Token Plan route |
| Standard pay-as-you-go price | No public per-million-token Qwen3.8 price was listed at verification time |
| Downloadable weights and license | Not published at verification time |
Numbers described as integration settings apply to the cited QwenCloud client configuration. They should not automatically be assigned to Qwen Studio, Qoder, Fireworks AI or another provider.
What is Qwen3.8 Max Preview?
Qwen3.8 Max Preview is the currently documented hosted preview of Qwen3.8. The official Qwen announcement says the preview debuted through Alibaba’s Token Plan, Qoder and QoderWork, and that Qwen3.8 is intended to become open weight. The current QwenCloud model directory labels the exact ID qwen3.8-max-preview as “Token Plan only.”
Qoder describes the model as having 2.4T parameters and positions it for software engineering, professional productivity, full-stack development, data analysis and Office workflows. Those performance descriptions are vendor claims, not independent benchmark conclusions. No public Qwen3.8 benchmark package with enough detail to reproduce a universal ranking was found in the official materials reviewed for this page.
Qwen3.8 is not Qwen3-8B
The names are easy to confuse, but they refer to different models:
| Name | What it means |
|---|---|
qwen3.8-max-preview | A hosted preview in the newer Qwen3.8 generation; Qwen/Qoder report 2.4T total parameters |
qwen3-8b | A separate Qwen3 checkpoint whose “8B” denotes approximately eight billion parameters |
Do not use “Qwen 3.8B” as an alternative name for Qwen3.8 Max Preview. It would incorrectly turn a version number into a parameter count and could mislead search users looking for Qwen3-8B.
What does “preview” mean?
QwenCloud states that the model’s capabilities will continue to improve during the preview period. It also says that, after the preview ends, the model may be taken offline or replaced by a production version. This has practical consequences:
- Behavior can change: output quality, latency and tool behavior may evolve without a new public snapshot ID.
- The alias is not guaranteed to be permanent: applications need a tested fallback and a migration plan.
- Results may not be reproducible indefinitely: record the model ID, date, product and settings used for every evaluation.
- Promotions are temporary: a preview Credits discount is not a permanent API price.
- Production assumptions are unsafe: do not infer a service-level commitment, frozen behavior or long-term availability from preview access alone.
Confirmed capabilities and limits
2.4T total parameters
Qwen and Qoder describe Qwen3.8 Max Preview as a 2.4-trillion-parameter model. This is a total parameter figure reported by the vendor. The public sources reviewed here do not disclose whether the architecture is dense or mixture-of-experts, how many parameters are active per token, or how the model is divided into layers and experts. Those values should remain unpublished until an official model card supplies them.
Reasoning and thinking behavior
QwenCloud lists Thinking support for the preview. Its official Qwen Code and OpenCode Token Plan configurations run this model with thinking enabled. Controls are product-specific: Qoder separately documents a thinking toggle in its model interface. Follow the current guide for the client you actually use instead of assuming that one product’s toggle or effort labels apply everywhere.
Visual understanding and text output
QwenCloud’s Personal and Team Token Plan tables list visual understanding and text generation for qwen3.8-max-preview. This supports image-based analysis through documented integrations. It does not mean that the model itself generates images or videos; those are separate models and platform tools. Do not claim general audio or video understanding unless the exact product documentation confirms it.
1M context and the exact client configuration
QwenCloud’s text-model table places Qwen3.8 Max Preview in the 1M context category. Its official OpenCode configuration uses a contextWindow of 983616 and maxOutputTokens of 131072. These are published client settings, not proof that every provider exposes identical limits. Leave room for output, thinking and tool results, and follow the error limits returned by the active endpoint.
Function Calling, built-in tools and structured output
The current QwenCloud model table marks Function Calling and built-in tools as supported. It does not mark structured output as supported for this preview. Function Calling still requires the surrounding application to validate arguments, enforce permissions, execute the function and return the result; the model does not independently perform a private application action.
Token Plan Harness tools
QwenCloud’s Harness documentation lists the following tools for Qwen3.8 Max Preview in both Personal and Team editions:
- Web search
- Code interpreter
- Web scraping
- Reverse image search
- Text-to-image search
These are platform services around the model. They are not proof that the base model natively browses the web, executes Python or generates images. “Text-to-image search” retrieves relevant images; it is not text-to-image generation. Successful Harness calls also consume plan Credits.
Where is Qwen3.8 Max Preview available?
| Product or provider | Status on July 26, 2026 | What can safely be said |
|---|---|---|
| QwenCloud Token Plan Personal | Confirmed | The exact model ID appears in the official allowlist |
| QwenCloud Token Plan Team | Confirmed | The exact model ID appears in the official allowlist |
| Qwen Code with Token Plan | Confirmed | The official QwenCloud setup includes qwen3.8-max-preview |
| Qoder | Confirmed | Qoder documents the model and current plan eligibility |
| QoderWork | Confirmed announcement | The official Qwen launch announcement names QoderWork; check current product access |
| Standard QwenCloud pay-as-you-go API | Not listed for this ID | QwenCloud currently labels the preview “Token Plan only” |
| Qwen Studio default model | Not confirmed | The public interface displayed Qwen3.7-Plus when checked; this does not prove Qwen3.8 is unavailable in every account |
| Fireworks AI Serverless | Not listed when checked | Fireworks’ public Serverless catalog listed Qwen3.7 Plus but not Qwen3.8 |
| Official weight download | Not released when checked | No official Qwen3.8 checkpoint and license were linked by the launch materials |
Availability is time-sensitive. Recheck the public Qwen Studio interface, the Fireworks Serverless catalog and the official QwenCloud model list before updating any availability claim.
How to access it with Qwen Code
The simplest official route is to open Qwen Code, enter /auth, choose the QwenCloud Token Plan and provide the dedicated Token Plan key. Then select the exact model ID qwen3.8-max-preview. If it is absent, update the client and check the model allowlist for your subscription.
For advanced configuration, QwenCloud publishes an OpenAI-compatible Token Plan route for its Singapore/Global service. The following is a shortened Qwen Code configuration based on the official guide:
{
"env": {
"BAILIAN_TOKEN_PLAN_API_KEY": "YOUR_TOKEN_PLAN_KEY"
},
"modelProviders": {
"openai": [
{
"id": "qwen3.8-max-preview",
"name": "Qwen3.8 Max Preview",
"baseUrl": "https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1",
"envKey": "BAILIAN_TOKEN_PLAN_API_KEY",
"generationConfig": {
"extra_body": {
"enable_thinking": true
}
}
}
]
}
}This is a Qwen Code configuration, not a general backend API example. Token Plan keys, Coding Plan keys and pay-as-you-go keys use different Base URLs and are not interchangeable. Never paste a real key into a web page, repository, screenshot or client-side application.
Region, usage and data terms
- Region: QwenCloud’s current Token Plan terms identify Singapore as the available region and Global as the deployment mode. They warn that prompts and outputs involve cross-border data transfer.
- Interactive use only: Personal and Team Token Plan documentation limits the service to compatible interactive programming and agent tools. Automated scripts, custom application backends and non-interactive batch use are prohibited.
- Data training: QwenCloud’s Token Plan FAQ states that conversation data is not used to train models.
- Credential separation: Token Plan, Coding Plan and pay-as-you-go keys and Base URLs cannot be mixed.
- Regional documentation: Alibaba’s mainland-China documentation uses different China (Beijing) endpoints. The configuration above is for QwenCloud’s Singapore/Global Token Plan; use the endpoint shown in your own console and regional guide.
These terms affect whether the preview fits a real workload. A compatible protocol does not override the subscription’s permitted-use restrictions.
Qwen3.8 Max Preview pricing
Qwen3.8 Max Preview currently uses Token Plan Credits, not a stable public input/output price per one million tokens. QwenCloud says Credits per request depend on the model, token usage, thinking and tool calls. Personal quotas use 5-hour and 7-day windows; Team quotas use a monthly seat model.
The official Token Plan pages currently advertise a preview promotion in which Qwen3.8 consumption can be as low as 10% of the standard Credits rate. Personal Edition also advertises an additional night discount for eligible use between 22:00 and 08:00 (UTC+8). QwenCloud explicitly reserves the right to change these offers, so they should not be presented as permanent pricing.
Qoder’s promotional coefficient and QwenCloud Token Plan Credits are separate billing systems. Do not convert either one into an invented per-token price. For established pay-as-you-go models, see this site’s Qwen API pricing guide; it should not be used to assign a price to qwen3.8-max-preview.
Qwen3.8 Max Preview vs Qwen3.7 Max
Qoder positions Qwen3.8 Max Preview as an improvement over Qwen3.7 Max for coding and professional workflows. Without a public, reproducible Qwen3.8 benchmark package, the safest comparison is based on documented product status and features rather than a universal performance ranking.
| Item | Qwen3.8 Max Preview | Qwen3.7 Max |
|---|---|---|
| Model ID | qwen3.8-max-preview | qwen3.7-max |
| Lifecycle | Changing preview | Regular QwenCloud model with a rolling ID and documented snapshots |
| QwenCloud context category | 1M | 1M |
| Token Plan capability table | Reasoning, visual understanding, text generation | Reasoning, text generation |
| Function Calling | Listed | Listed |
| Built-in tools | Listed | Listed |
| Structured output in current table | Not listed | Not listed |
| Documented QwenCloud route | Token Plan only | Regular API and supported plans |
| Best fit today | Controlled evaluation of the newest preview | Workloads that need a regular documented model lifecycle |
Do not migrate only because 3.8 is newer. Test both IDs with the same prompts, tool definitions, visual inputs, context lengths and success criteria. Record the date and access route because preview behavior can change between evaluations.
What has not been officially confirmed?
The official sources reviewed for this page do not yet provide a complete production model card. The following details should not be published as facts:
- Dense or mixture-of-experts architecture
- Activated parameter count, layer count or expert configuration
- Training-token count, training-data composition or knowledge cutoff
- A model-specific language count
- A frozen production ID, release date or service-life commitment
- A public checkpoint repository, quantization list or final license
- Self-hosting hardware requirements or fine-tuning support
- A permanent pay-as-you-go input/output token price
- Identical context, output and thinking controls on every provider
- Independent benchmark leadership over named competing models
Qwen’s statement that Qwen3.8 is “going open-weight soon” is a future plan, not a completed release. Until an official repository, weights and license are published, the current preview should be described as hosted—not open source, downloadable or self-hostable.
Should you use Qwen3.8 Max Preview?
It is reasonable to evaluate the preview if you use QwenCloud Token Plan or Qoder and want to test reasoning, coding, visual analysis or long-context work. Use representative private evaluations rather than relying on marketing language or one public score.
Before depending on it, confirm:
- Your account can select the exact
qwen3.8-max-previewID. - Your dedicated Token Plan key matches the correct regional Base URL.
- Your use is interactive and permitted by the plan terms.
- Your application can tolerate preview changes, thinking latency and Credits consumption.
- Your prompts leave sufficient room for thinking, tool results and final output.
- You have tested the actual image formats, functions and Harness tools your workflow requires.
- Your privacy review covers Global deployment and cross-border data transfer.
- You have a tested fallback if the preview is updated, withdrawn or replaced.
Frequently asked questions
What is Qwen3.8 Max Preview?
It is a hosted, evolving Qwen model available under the exact ID qwen3.8-max-preview. QwenCloud currently documents it through Token Plan, while Qwen’s announcement also names Qoder and QoderWork.
Is Qwen3.8 the same as Qwen3-8B?
No. Qwen3.8 is a newer version name. Qwen3-8B is a separate earlier checkpoint with approximately eight billion parameters. Qwen/Qoder report 2.4 trillion total parameters for Qwen3.8 Max Preview.
Is Qwen3.8 Max fully released?
No. The accessible model is explicitly labeled Preview. QwenCloud says it may be iteratively improved, taken offline or replaced by a production version.
What is the exact Qwen3.8 model ID?
The exact current ID is qwen3.8-max-preview. QwenCloud treats its supported-model table as an exact-string, case-sensitive allowlist.
Where can I access Qwen3.8 Max Preview?
Confirmed routes include QwenCloud Token Plan Personal and Team, Qwen Code configured with Token Plan, Qoder and QoderWork. Product eligibility can depend on the current account, plan and client version.
Is Qwen3.8 Max Preview the default model in Qwen Studio?
No official source reviewed for this page identifies it as the default. The public Qwen Studio interface showed Qwen3.7-Plus on July 26, 2026. That observation does not prove that the preview is unavailable to every account.
Is Qwen3.8 Max Preview available on Fireworks AI?
It was not listed in Fireworks’ public Serverless catalog when checked. Fireworks availability must be verified through Fireworks’ own live catalog; Alibaba or QwenCloud access does not prove third-party availability.
Does Qwen3.8 Max Preview have a 1M-token context window?
QwenCloud lists it in the 1M context category. Its OpenCode example uses a 983,616-token context setting and a 131,072-token maximum-output setting. Treat those as QwenCloud integration values, not universal limits for every product.
Does it support images or video?
Visual understanding and image input are documented. A general video-input claim is not included here because the model-specific Token Plan capability table does not list it. Image and video generation require separate models or platform tools.
Can I disable thinking?
QwenCloud’s documented Qwen Code and OpenCode Token Plan integrations enable thinking and describe it as always on for this preview. Qoder documents a product-specific thinking toggle. Check the current guide for your exact client instead of assuming one universal control.
Can I use Token Plan for a custom application backend?
No. QwenCloud’s current Token Plan terms restrict both Personal and Team editions to interactive use in compatible programming and agent tools. Automated scripts, application backends and non-interactive batch use are prohibited.
Is Qwen3.8 Max Preview open source or downloadable?
Not at the time of verification. Qwen says an open-weight release is planned, but no official Qwen3.8 checkpoint, repository or license had been published. Do not download unofficial files presented as Qwen3.8 without verified provenance and license terms.
What is the Qwen3.8 API price?
No standard public per-million-token price was listed for this preview. Current access uses Token Plan Credits or separate Qoder Credits, and promotions can change. Check the active console before purchase or use.
Is Qwen3.8 Max Preview production-ready?
It should be treated as a preview, not a frozen production model. A production team should use controlled evaluation, monitoring, a fallback and a migration plan, and should confirm that the plan’s interactive-use restriction fits the intended workflow.
How this page was verified
This page prioritizes first-party Qwen, QwenCloud, Alibaba Cloud and Qoder materials. Changing availability claims were checked against the public Qwen Studio interface and Fireworks Serverless catalog. Vendor performance statements are attributed, unsupported fields are left undisclosed, and no claim of independent hands-on model testing is made.
Update log: July 26, 2026 — first publication; verified preview status, exact ID, 2.4T vendor figure, QwenCloud Token Plan access, 1M context category, client configuration, Harness tools, usage restrictions, Qwen Studio public default and Fireworks Serverless listing.
Official sources
- Qwen official launch announcement
- Qwen official 2.4T follow-up
- QwenCloud Token Plan overview
- QwenCloud Token Plan Personal Edition
- QwenCloud Token Plan Team Edition
- QwenCloud Token Plan FAQ and restrictions
- QwenCloud text-generation model directory
- QwenCloud Qwen Code configuration
- QwenCloud OpenCode configuration and limits
- QwenCloud Harness tools
- Qoder model selection documentation
- Qoder Qwen3.8 Max Preview release and promotion
- Official Qwen organization on Hugging Face
Continue exploring Qwen
Review established Qwen releases and provider-specific API requirements before choosing a model for production.
Editorial maintenance note: Recheck this page when Qwen publishes a production Qwen3.8 model card, official weights, a license, a dated snapshot, standard API pricing or a change to Token Plan availability.