Qwen3.6-Max-Preview: Alibaba's New Coding-Preview Tier Above Qwen3.6-Plus
Qwen3.6-Max-Preview is the new preview tier above Qwen3.6-Plus, launched on April 20, 2026 through Qwen Studio and Alibaba Cloud Model Studio. The current public source trail is good, but it is not all one kind of source. The launch post is strongest as a benchmark delta sheet: SkillsBench +9.9, SciCode +10.8, NL2Repo +5.0, Terminal-Bench 2.0 +3.8, SuperGPQA +2.3, and ToolcallFormatIFBench +2.8 over Plus, plus six #1 benchmark claims. The pricing docs are strongest as route evidence: the public Model Studio page already separates Mainland and International pricing for `qwen3.6-max-preview`, while the Global (Virginia) section still emphasizes the older `qwen3-max` family. That route split is the part most articles miss.
- The official launch post frames Qwen3.6-Max-Preview as the preview tier above Qwen3.6-Plus, not as a renamed Plus model.
- Six #1 launch claims: SWE-bench Pro, Terminal-Bench 2.0, SkillsBench, SciCode, QwenClawBench, and QwenWebBench.
- The strongest public proof today is the launch-post delta sheet, not a full standalone model card.
- Official pricing is already route-specific: Mainland and International rows publish `qwen3.6-max-preview`, while the Global section still emphasizes `qwen3-max` routes.
What Alibaba actually launched on April 20, 2026
Qwen3.6-Max-Preview first appeared in the official Qwen WeChat launch post on April 20, 2026. The launch positioning is straightforward: this is the early preview of the next flagship tier above Qwen3.6-Plus, with a strong emphasis on agentic coding, world knowledge, and instruction following.
That wording matters because the public evidence is still launch-shaped. The best way to use the current sources is to quote the launch-post deltas and the route-specific pricing pages, then explain clearly that this is a preview route still being iterated. That is more accurate than pretending a full standalone model card already exists.
- Launch date confirmed in the official Qwen WeChat article: April 20, 2026.
- Preview route name confirmed in the launch post and pricing page: `qwen3.6-max-preview`.
- Availability confirmed in the launch post: Qwen Studio first, Model Studio API next.
- The launch post explicitly says the preview is still being iterated and later versions will keep improving.
Official image
Alibaba is still framing Qwen3.6-Max-Preview as a launch-stage preview
The official WeChat announcement is still the clearest public image source for this model. It reinforces the key positioning point: Qwen3.6-Max-Preview is an early preview tier above Plus, not a renamed Plus row.
- Useful when readers need visual proof that the model name includes the `Preview` suffix.
- Pairs naturally with the route-specific pricing rows on the Model Studio side.
The launch post is strongest as a benchmark delta sheet
The cleanest public benchmark takeaway today is not one giant table. It is the set of launch-post deltas over Qwen3.6-Plus. That is what gives you a credible, source-backed way to say the preview is materially above Plus on coding and tool-heavy work.
The post also claims six benchmark championships. That is useful, but it should be quoted as a launch claim from the official Qwen source rather than rewritten as if it were an independent external leaderboard with one reusable table.
Official image
Alibaba uses an inline benchmark poster to show the jump from Plus to Max-Preview
This image is embedded directly inside the official launch article. It is the strongest public visual proof for the benchmark-delta story around Qwen3.6-Max-Preview before a fuller standalone model card appears.
- Useful when the article needs one official image for the “six #1 claims plus deltas over Plus” framing.
- Pairs naturally with the route-specific Model Studio pricing rows because both are still launch-stage public sources.
| Benchmark or signal | What the official launch says | How to write it safely |
|---|---|---|
| SWE-bench Pro | #1 win | Use it as a rank claim from the launch post, not as a full external benchmark table |
| Terminal-Bench 2.0 | #1 win and +3.8 vs Plus | You can cite the rank and the delta; avoid inventing unrelated model-card fields |
| SkillsBench | #1 win and +9.9 vs Plus | One of the cleanest coding-improvement claims in the whole post |
| SciCode | #1 win and +10.8 vs Plus | Strong evidence for scientific and code-heavy tasks |
| QwenClawBench / QwenWebBench | #1 wins | These are Qwen-owned benchmarks; label them that way |
| Artificial Analysis board | Best Chinese model according to the launch post | Treat it as a leaderboard claim and recheck live if you need the current exact rank |
These are the clearest quantitative claims in the official launch post, and they are the safest benchmark facts to reuse in an article today.
Official Qwen launch post.
Official Qwen launch post.
Official Qwen launch post.
Official Qwen launch post.
Official Qwen launch post.
Official Qwen launch post.
The launch post also cites QwenChineseBench +5.3. QwenClawBench and QwenWebBench are listed as #1 wins but without the same reusable delta format. Source: Official Qwen3.6-Max-Preview WeChat announcement.
Model Studio pricing already shows route differences
The official Model Studio pricing page is currently the best route document for this preview model. It shows that `qwen3.6-max-preview` is not simply a generic “Max” row copied to every region. Mainland and International (Singapore) currently publish explicit `qwen3.6-max-preview` pricing, while the Global (Virginia) section still emphasizes older `qwen3-max` and `qwen3-max-preview` rows.
That means a buyer-facing article should not present one universal overseas price for the preview model. The route scope itself is already different on the official page.
- The current public pricing bands top out at 256K context. That is the safest published context number to use today.
- The public pricing page currently distinguishes `qwen3.6-max-preview` from `qwen3-max` and `qwen3-max-preview` rather than folding them into one “Max” family row.
- If your article needs a single plain-English buying rule, write the route scope first and the token price second.
| Deployment scope | Context band | Input price | Output price | Notes |
|---|---|---|---|---|
| Mainland China | 0-128K | 9 CNY / 1M input tokens | 54 CNY / 1M output tokens | The pricing page also shows a mainland free-quota note for the first 90 days of account life. |
| Mainland China | 128K-256K | 15 CNY / 1M input tokens | 90 CNY / 1M output tokens | Use this row for the longer context band. |
| International (Singapore) | 0-128K | 9.742 CNY / 1M input tokens | 58.455 CNY / 1M output tokens | Published in the international section of the same pricing page. |
| International (Singapore) | 128K-256K | 14.988 CNY / 1M input tokens | 89.93 CNY / 1M output tokens | Published separately from the mainland route. |
Protocol support, Anthropic compatibility, and preserve_thinking
The launch post says the preview supports `preserve_thinking`, specifically to keep prior-round thinking content in the message history for agent-style workflows. That is a model-specific launch detail rather than a generic protocol statement, so it is worth citing directly from the launch source.
For transport and SDK compatibility, Alibaba's public Anthropic-compatible documentation is the cleanest source. It explicitly publishes Beijing and Singapore Anthropic-style base URLs under the DashScope / Model Studio route. That gives readers a concrete path for using the preview through an Anthropic-style client, even when the model-specific example is not written for this exact preview tier.
- Model name to watch for in the API route: `qwen3.6-max-preview`.
- The public launch post recommends `preserve_thinking` for agent tasks.
- The current public docs prove protocol compatibility and route scopes more clearly than they prove every higher-level model feature claim.
| Scope | Base URL | What is confirmed |
|---|---|---|
| Mainland China | https://dashscope.aliyuncs.com/apps/anthropic | Anthropic-compatible Messages API route is documented publicly. |
| International (Singapore) | https://dashscope-intl.aliyuncs.com/apps/anthropic | Anthropic-compatible Messages API route is documented publicly. |
How to position it in a real guide right now
| Question you are answering | Best row to cite today | Why |
|---|---|---|
| I need a stable public Qwen benchmark row | Qwen3.6-Plus | The older Qwen3.6 release page is still the cleaner full benchmark object. |
| I need the newest Qwen frontier preview above Plus | Qwen3.6-Max-Preview | The launch post provides the delta sheet that proves the preview moved up meaningfully. |
| I need current published token pricing by route | Model Studio pricing page | That is where the mainland and international scope split is documented. |
| I need one universal overseas Qwen Max price | Do not do this | The official pricing page is already route-specific, so flattening it loses information. |
Use Qwen3.6-Max-Preview as a route-aware upgrade above Plus
The cleanest article framing today is simple: cite the launch-post gains over Plus, cite the route-specific pricing rows, and avoid inventing one flattened “global Max” story that the official docs do not publish.
Sources and official links
Frequently asked questions
Why is it called Qwen3.6-Max-Preview instead of Qwen3.6-Max?
Because Alibaba is explicitly publishing it as a preview-stage flagship tier. The preview suffix is part of the public model name on the launch post and the pricing page, and the launch article says later versions will keep being improved.
How does Qwen3.6-Max-Preview differ from Qwen3.6-Plus?
The public evidence right now is mostly comparative. The launch post positions Max-Preview above Plus with six #1 benchmark claims and deltas such as SkillsBench +9.9, SciCode +10.8, NL2Repo +5.0, and Terminal-Bench 2.0 +3.8 over Plus. It also gets its own preview pricing rows on part of the Model Studio pricing page.
Is 1M context confirmed for this preview model?
No public source in this guide confirms a 1M context window for `qwen3.6-max-preview`. The current public pricing rows top out at 256K context bands, so 256K is the safest published number to use until Alibaba releases a fuller model card.
Is multimodal support confirmed for Qwen3.6-Max-Preview?
The current launch post and pricing docs are much clearer about coding, reasoning, instruction following, and API protocol compatibility than they are about modality support. Until Alibaba publishes a dedicated model page, the safe framing is “reasoning and coding preview model” rather than “full multimodal flagship.”
Is Qwen3.6-Max-Preview available outside mainland China?
Yes, but the public route story is specific rather than universal. The Model Studio pricing page publishes international pricing for Singapore, and the Anthropic-compatible docs publish a Singapore base URL. That is different from saying every global scope exposes the same preview row today.