Alibaba Qwen Models: Every Model, Spec, Price and Release Date

Qwen is the large language model family from Alibaba's Qwen Team, offered as paid APIs on Alibaba Cloud Model Studio and as open weights on Hugging Face, and it moved through four generations, Qwen3.5 to Qwen3.8, in 2026. This hub lists the Qwen models in Alibaba Cloud's public documentation and the Qwen Hugging Face model cards as of 15 November 2026, with API IDs, release dates, context windows, prices, licences, availability and deprecation dates. It is written for developers, buyers, architects and anyone who needs to know which Qwen model is which. Models are listed by release date, newest first, an objective order that implies no ranking. Every fact comes from Alibaba Cloud's documentation or the official Qwen Hugging Face cards, cited in the sources list.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It uses Alibaba Cloud Model Studio's public documentation as of November 2026. Model details change, so confirm against Alibaba Cloud's current documentation.

Key takeaways

• Qwen3.8 is the latest generation. The API flagship is qwen3.8-max, with qwen3.8-flash and qwen3.8-omni-flash alongside, and qwen4-max-preview is listed under Alibaba Cloud's Token Plan.
• Context reaches 1M tokens. Open Qwen3.5 to Qwen3.8 weights list 262,144 tokens natively, extensible to about 1M, and the hosted Max and Flash versions use 1M by default.
• Prices run from $0.03 to $2.50 per million input tokens. On the Singapore deployment qwen3.7-flash starts at $0.030 input for prompts up to 32K, and qwen3.7-max is $2.50 input and $7.50 output; qwen3.8-max is $2 and $6.
• Licences differ by model. Qwen3.8-27B, Qwen3.6 and Qwen3.5-397B-A17B use Apache 2.0; Qwen3.8-2.4T-A95B uses a qwen3.8-max licence and Qwen3.8-Flash-Next qwen-community-1.0.
• Deprecations run in batches. Model Studio's deprecation page lists batches deprecated on 10 October 2026 and one scheduled for 19 January 2027.

What is Alibaba Qwen?

Qwen is developed by the Qwen Team at Alibaba Group. Alibaba Cloud serves the hosted Qwen models through Model Studio, and the model cards also refer to Qwen Cloud as a serving channel. Qwen uses two parallel naming schemes: API tiers (Max, Plus, Flash, Omni) with dated snapshot IDs such as qwen3.7-plus-2026-05-26, and open-weight checkpoints named by parameter count, such as Qwen3.8-27B or Qwen3.5-397B-A17B, where A17B means 17 billion active parameters. The model cards describe several hosted models as the official versions of open checkpoints with added production features.

The main table lists the Qwen3.5 to Qwen3.8 models named in this page's scope, plus qwen4-max-preview. Facts are as of 15 November 2026.

Current Qwen models at a glance

As of 2026-11-15. Source: Model Studio models and pricing pages and Qwen Hugging Face cards. Models are listed by release date, newest first, with undated entries first. Category is the Qwen tier or checkpoint type.

Where the context column says "pricing tier", the figure comes from the highest input-token band on the pricing page. The Model Studio models page describes qwen3.8-omni-flash as handling text, images, audio and video. See Open-Weight vs Closed Models Explained.

Key facts for the main Qwen models

As of 2026-11-15. Identical fields for every model; "Not stated by Alibaba Cloud" means not stated by Alibaba Cloud or the Qwen Team on the pages checked (2026-11-15). Release month is from the model card citation where no exact date is published.

The Qwen3.8-2.4T-A95B card states that thinking mode cannot be disabled on the open weights and that responses begin with a reasoning block; the hosted Qwen3.8-Max adds non-thinking support, vision input, a 1M default context and built-in tools. Its architecture is a 92-layer hybrid of Gated DeltaNet and Gated Attention layers with 512 experts, 10 routed and 1 shared active per token. See Mixture of Experts Explained.

Features stated across current models

• Thinking modes: the pricing page lists separate non-thinking and thinking output prices for Plus models, and qwen3.7-max-preview is thinking mode only.
• Context cache: listed for Max, Plus and Flash models; the pricing page gives explicit cache creation at 125% and cache hits at 10% of the standard input price as an example.
• Batch: 50% off the real-time price where listed, not combinable with the cache discount.
• Omni and realtime: qwen3.8-omni-flash supports Chat Completions or Responses, and qwen3.8-omni-flash-realtime handles real-time audio and video conversation.
• Serving frameworks: the Qwen3.8-2.4T-A95B card lists compatibility with SGLang, vLLM and TokenSpeed.

Qwen model families and naming

• Max: the hosted flagship tier (qwen3.7-max, qwen3.8-max) and the legacy qwen-max, plus qwen4-max-preview.
• Plus: the mid tier, which the models page describes as balancing performance, speed and cost (qwen3.5-plus, qwen3.6-plus, qwen3.7-plus).
• Flash: the fast, low-cost tier (qwen3.5-flash to qwen3.8-flash).
• Omni: multimodal models with audio and video input and speech output (qwen3.5-omni-plus, qwen3.8-omni-flash).
• Open checkpoints: named by size, for example Qwen3.8-27B (dense) and Qwen3.6-35B-A3B (mixture of experts, 3B active), with FP8 variants.
• Snapshot IDs: an undated API name points to a dated snapshot, for example qwen3.7-plus currently resolves to qwen3.7-plus-2026-05-26. See LLM Naming and Versioning Explained.

When was each Qwen model released?

As of 2026-11-15. Alibaba Cloud does not publish a single release-notes timeline on the pages checked. Dates below are API snapshot dates from the pricing page and release months from Hugging Face card citations, in date order.

The latest generation is Qwen3.8, whose card citations are dated August 2026; the qwen3.8-max-0902 snapshot is also listed on the pricing page. Qwen4-Max-Preview has no published release date on the pages checked.

How much do Qwen models cost?

As of 2026-11-15. Price per million tokens on the Singapore (International) deployment, shown with a "$" sign on the pricing page (the page does not name the currency code). Cached input is the 10% cache-hit example on the pricing page; cache write is the 125% creation example. Source: Model Studio pricing page.

• Free quota: 1M free tokens on qwen3.8-max, qwen3.8-flash, qwen3.7-max, qwen3.7-plus and qwen3.7-flash, valid for 90 days and only in the Singapore region.
• Tiered pricing: several models charge by input-token band, so long prompts cost more per token.
• Cache and batch: the two discounts cannot be combined; the Context Cache page sets the detailed rules.
• Other regions: mainland China prices are listed separately and are not reproduced here.
• Not listed: qwen4-max-preview and qwen3.8-omni-flash prices are not stated on the portion of the pricing page checked.

See Model Pricing Explained and How to set spend limits on LLM API keys.

Where are Qwen models available?

• First-party: Alibaba Cloud Model Studio, with Qwen Cloud named on the model cards as the serving channel for Qwen3.8-Max and Qwen3.8-Flash.
• Regions: the pricing page lists Singapore (International), US (Virginia), Germany (Frankfurt), Hong Kong (China), Japan (Tokyo) and China (Beijing, Chinese mainland). Model availability can differ by region.
• Open weights: the Qwen organisation on Hugging Face hosts Qwen3.8, Qwen3.6 and Qwen3.5 checkpoints, with FP8 variants.
• Cloud platforms: availability on other providers' clouds is not stated by Alibaba Cloud on the pages checked (2026-11-15).

Open checkpoints can be self-hosted, which moves operation, patching and access control to the organisation. See the Self-Hosted LLM Security Guide.

Restricted-access Qwen models

As of 2026-11-15. Source: Model Studio models and pricing pages.

No Qwen model is described as restricted to verified or invited organisations on the pages checked.

What is the context window of Qwen models?

The open Qwen3.5, Qwen3.6 and Qwen3.8 checkpoints list 262,144 tokens natively, extensible with YaRN scaling to 1,010,000 tokens (1,000,000 on Qwen3.8-27B and Qwen3.8-Flash-Next). The Qwen3.6-27B card notes that static YaRN scaling may affect performance on shorter texts. The hosted Qwen3.5-Plus, Qwen3.8-Max and Qwen3.8-Flash use 1M tokens by default, per their cards, and the Plus and Flash pricing tiers run up to 1M input tokens. Maximum output tokens are not stated on the pages checked. See Context Windows Explained.

Is any Qwen model open source?

The Qwen Team publishes weights for many models. Qwen3.8-27B, Qwen3.6-35B-A3B, Qwen3.6-27B and Qwen3.5-397B-A17B are released under Apache 2.0. Qwen3.8-2.4T-A95B, the base of Qwen3.8-Max, carries a licence named qwen3.8-max, and Qwen3.8-Flash-Next carries qwen-community-1.0; read each repository's licence file before use. The hosted Max, Plus, Flash and Omni API models, and qwen4-max-preview, are API only. Training data is not published. See the Open-Weight Model Security Guide.

Model cards and system cards

As of 2026-11-15.

See Model Cards and System Cards Explained.

Benchmarks

As of 2026-11-15. Provider-reported and independent results are kept in separate columns. Scores are as the Qwen Team reports them on the Hugging Face cards.

The Qwen3.8-2.4T-A95B card reports these scores in a Qwen3.8-Max column and does not state whether they apply to the open weights or the hosted version. See How to Read LLM Benchmarks.

Deprecations and retirements

As of 2026-11-15. Source: Model Studio model deprecation page. Model IDs in each batch are given in the linked announcements, which this page does not reproduce.

The deprecation page points to the Singapore region's Model Monitoring page to check whether an account still calls a retiring model. See the AI Model Retirement Guide.

Data handling and security

• Training use: the Model Studio security and privacy page (updated 11 October 2026) states that Alibaba Cloud will never use customer data for model training.
• Retention: the page states that Model Studio stores data generated from model and application calls; retention terms are set in the related agreements.
• Encryption: data transmitted during application building and model training is encrypted with AES-256, per the same page.
• Assurance: the page states that Model Studio has achieved SOC 2 compliance with an unqualified opinion.
• Region choice: deployments are listed in six regions, including the Chinese mainland, so the chosen region determines where calls are served.
• Key handling: generic controls apply to every provider's API: keep keys out of code, scope and rotate them, and set spend limits.

See AI Data Retention and Deletion and the Alibaba Cloud Security Guide.

How to choose a Qwen model: factual criteria

• Tier: Max, Plus and Flash differ in price per token; Omni models add audio and video.
• Price per token: compare tiered input bands, output rates and the batch and cache discounts.
• Region: confirm the model is offered in the chosen deployment region; free quota applies only in Singapore.
• Snapshot pinning: call a dated snapshot ID where reproducible results matter, since undated names move.
• Hosting model: open checkpoints can run in the organisation's own environment; hosted Max adds vision, non-thinking mode and built-in tools over the Qwen3.8 weights.
• Licence: Apache 2.0 for Qwen3.8-27B, Qwen3.6 and Qwen3.5-397B-A17B; separate licences for Qwen3.8-2.4T-A95B and Qwen3.8-Flash-Next.
• Hardware for self-hosting: sizes range from 27B dense to 2.4T total parameters.
• Lifecycle: check the deprecation batches before building on an older snapshot.

Practitioner checklist

• Pin production calls to a dated snapshot ID and record it in the AI inventory.
• Check the Model Studio deprecation page and Model Monitoring before 19 January 2027.
• Choose the deployment region deliberately and document it for data protection records.
• Re-estimate cost when prompts cross a pricing band such as 256K tokens.
• Read the licence file of each open checkpoint before redistribution or commercial use.
• Store Model Studio API keys in a secrets manager and set spend limits. See the LLMjacking Guide.

Standards and references

• Alibaba Cloud: Model Studio models (accessed 2026-11-15)
• Alibaba Cloud: Model Studio model pricing (accessed 2026-11-15)
• Alibaba Cloud: Model Studio model deprecation (accessed 2026-11-15)
• Alibaba Cloud: Model Studio security certifications and privacy (accessed 2026-11-15)
• Qwen Team: Qwen3.8-2.4T-A95B model card (accessed 2026-11-15)
• Qwen Team: Qwen3.8-Flash-Next model card (accessed 2026-11-15)
• Qwen Team: Qwen3.8-27B model card (accessed 2026-11-15)
• Qwen Team: Qwen3.6-35B-A3B model card (accessed 2026-11-15)
• Qwen Team: Qwen3.5-397B-A17B model card (accessed 2026-11-15)

Related NHI Mgmt Group resources: Reasoning Models Explained, Multimodal Models Explained, Open-Weight Model Security Guide, LLM Security Guide.