Qwen3.8: Everything We Know About Alibaba's 2.4T Model
Qwen3.8-Max-Preview is live right now, and Alibaba has published exactly zero benchmarks for it. The teaser claims 2.4 trillion parameters, an open-weight release, and a capability ranking second only to Claude Fable 5. The product pages that sell access to it mention the model by name in a single promotional line and then say nothing else at all.
I checked both Token Plan pricing pages directly. Qwen3.8-Max-Preview is confirmed as a real, purchasable product, so this is not vaporware. What does not exist yet is a parameter count, a context window, a benchmark table, a model card, per-token pricing, or an official Qwen launch post. The most recent flagship Alibaba has formally documented is still Qwen3.7-Max from May 2026.
STATUS AS OF JULY 19, 2026
CONFIRMED: Qwen3.8-Max-Preview exists and is being sold as early access through Alibaba Token Plan tiers. NOT CONFIRMED: the 2.4T parameter count, the open-weight promise, the Fable 5 ranking, and every performance figure. No official Qwen blog post, model card, or benchmark table has been published. This page is updated the moment official data lands.
Here is what this post gives you: a clean separation of confirmed facts from marketing claims, the verified Qwen lineage that tells you what Alibaba actually ships, an analysis of what 2.4 trillion parameters would mean against Kimi K3 and Claude Fable 5, and a checklist of what to watch before you move a workload. No invented benchmarks. If a number is not published, I say so.
What Is Qwen3.8?
Qwen3.8 is Alibaba's next flagship Qwen model, currently available as Qwen3.8-Max-Preview through Alibaba Token Plan, Qoder, and QoderWork. The teaser claims 2.4 trillion parameters and a coming open-weight release, positioning it second only to Claude Fable 5 among frontier models.
The preview build is real. Both Token Plan pricing pages list Qwen3.8-Max-Preview by name as an advanced model included in their subscription tiers, which is as close to official confirmation as exists today. What is missing is everything a buyer would actually need: Alibaba's Qwen team publishes a detailed launch post with full benchmark tables for each flagship, as it did for Qwen3.7 in May and Qwen3.6-Max-Preview in April, and no equivalent post exists for 3.8.
My read: the rollout fits Alibaba's established cadence of a new Max tier every four to six weeks in 2026, so a July Qwen3.8 following a May Qwen3.7 is exactly on schedule. Selling early access before publishing specs is a normal preview pattern. The 2.4T figure and the open-weight promise are the two parts I would not repeat as fact until Qwen publishes them.
If you want the verified picture of the model this replaces, start with our Qwen3.7-Max review, which covers the 1M-token agent capabilities in detail.
Verified vs Claimed: The Full Checklist
Here is every Qwen3.8 claim sorted by whether it can currently be verified. Bookmark this table; it is the fastest way to sanity-check anything else you read about this model in the next two weeks.
Quotable takeaway: Qwen3.8 is the rare launch where the product is confirmed real and every single performance claim about it is unconfirmed. You can buy access today and still not know what you are buying.
The claim I would scrutinise most: not the parameter count, but the open-weight promise, because Alibaba's own two most recent flagships both shipped closed.
The Qwen Lineage: How We Got to 3.8
Alibaba has shipped four flagship Qwen tiers in under a year, escalating from roughly 1 trillion parameters to a claimed 2.4 trillion. Understanding that cadence is the best available evidence for what Qwen3.8 will actually be, since the pattern has been remarkably consistent.
Two patterns matter here. First, every Max-tier flagship since Qwen3-Max has been proprietary, delivered through Alibaba Cloud Model Studio rather than Hugging Face. Second, Alibaba runs a genuine two-track strategy: the Max tier stays closed while the smaller Qwen3 and Qwen3-Coder lines ship under Apache 2.0, which is why Qwen3-Coder-480B sits in our open-model rankings at 69.6% SWE-bench Verified.
If Qwen3.8 truly goes open-weight at 2.4 trillion parameters, it breaks the Max-tier pattern completely. That would be the real story, bigger than the parameter count, because it would mean Alibaba decided to open-source its crown jewel in the same month Moonshot and Thinking Machines did the same.
We have covered every step of this ladder, from the Qwen3.6-Max-Preview review in April through the Qwen3.7 Max Preview arena results in May, which is the clearest record of how fast this line moves.
What Qwen3.7-Max Actually Delivers
Qwen3.7-Max is the verified baseline any Qwen3.8 claim should be measured against, and it is genuinely strong: 92.4 on GPQA Diamond, 80.4% on SWE-bench Verified, and 69.7 on Terminal-Bench 2.0-Terminus, with a 1 million token context window and a native extended-thinking mode.
Read that pricing row twice. Qwen3.7-Max delivers a 92.4 GPQA score at $1.25 input, which is one eighth of Claude Fable 5's input price and less than half of Kimi K3's. Alibaba's real weapon has never been topping a leaderboard, it is delivering 90% of frontier quality at 15% of frontier cost. Whatever Qwen3.8 turns out to be, expect that positioning to hold.
My honest take: if Qwen3.8 simply repeats the 3.7 formula with more parameters and keeps the pricing, it will be one of the best value models on the market without needing to beat anyone's benchmark.
Don't just use ChatGPT. Learn to build custom LLM agents, RAG pipelines, and full-stack Agentic AI apps in our intensive 6-week program.
What 2.4 Trillion Parameters Would Mean
A 2.4 trillion parameter Qwen3.8 would be the second largest model publicly known, behind Kimi K3's 2.8 trillion and ahead of Thinking Machines' 975 billion parameter Inkling. In a month where three labs pushed past a trillion, scale alone no longer buys headlines, but it does tell you something about architecture and cost.
The number that actually matters is not published anywhere: active parameters. Inkling runs 975 billion total but activates only 41 billion per token, which is why it can be served affordably. Kimi K3 and Qwen's Max tier are both sparse Mixture-of-Experts designs too. Without the active-parameter count, a 2.4T headline tells you nothing about whether you can afford to run it or whether any consumer hardware could host it.
The question I would ask Alibaba first: how many parameters activate per token? That single number decides whether an open-weight 2.4T model is a genuine gift to the community or a trophy almost nobody can serve.
Alibaba has already proven the point that size is not everything. Our review of Qwen3.6-27B beating a 397B model on coding is the best example of why active parameters and training quality matter more than a headline count.
The Open-Weight Question
The open-weight claim is the most consequential and least supported part of the Qwen3.8 announcement, because Alibaba's last two flagships shipped closed. Qwen3.7-Max and Qwen3.6-Max-Preview are both API-only through Alibaba Cloud Model Studio, with no weights published.
Alibaba does have a real open-source record, which is why the claim is not absurd. The Qwen3 and Qwen3.5 lines ship under Apache 2.0, one of the most permissive licenses available, and Qwen3-Coder-480B is a genuinely strong Apache 2.0 coding model. Alibaba has arguably done more for open AI than any Western lab except Meta, and Meta has now pivoted proprietary. So the company has both the history and the incentive.
What would make me believe it: a Hugging Face repository with an actual license file. Until that URL exists, an open-weight promise is a roadmap, and roadmaps slip. I have watched several 2026 launches promise weights within days and deliver weeks later, or in one case not at all. Moonshot promised Kimi K3 weights by July 27, and the community is watching that date the same way.
Practical rule: treat any promised open-weight release as vaporware until the repo exists, then verify the license file matches the announcement. Plan your architecture on what you can download today.
Where You Can Try Qwen3.8-Max-Preview
Access runs through three Alibaba surfaces: Token Plan, Qoder, and QoderWork. Qoder is Alibaba's agentic coding platform and Token Plan is its subscription bundle, sold in Lite, Standard, and Pro tiers with different credit allowances and concurrent-agent limits. Both Token Plan pages name Qwen3.8-Max-Preview as an included advanced model.
The Token Plan pages come in two versions, and they are not operated by the same entity. The international page at qwencloud.com lists its operator as Intelligent Cloud Computing (Singapore) Private Limited. The China page at platform.qianwenai.com presents as Alibaba's Qianwen AI Platform, Qianwen being the Chinese name for Qwen. Neither page publishes per-model pricing, only tier bundles.
CHECK THE OPERATOR BEFORE YOU PAY
Alibaba's best known Qwen surfaces are qwen.ai, Alibaba Cloud Model Studio at alibabacloud.com, and the Qwen Chat app. The international Token Plan page is operated by Intelligent Cloud Computing (Singapore) Private Limited rather than an alibabacloud.com domain. That may be a legitimate regional entity, but confirm it against your Alibaba Cloud account before entering payment details or API keys, and prefer the official console where you have the choice.
Safest path today: check the Alibaba Cloud Model Studio supported-models page for a Qwen3.8 entry and buy through your existing Alibaba Cloud account. If you are evaluating on a budget, the tier bundles make sense only once per-model pricing is published, since you cannot currently calculate cost per token.
Qwen3.8 vs Kimi K3 vs Fable 5: The Claim Examined
The claim that Qwen3.8 is second only to Claude Fable 5 cannot be evaluated, because no Qwen3.8 benchmark has been published. What we can do is check whether that ranking is even plausible given the verified field, and the honest answer is: possible on some benchmarks, unlikely across the board.
Look at the GPQA row. Qwen3.7-Max is already within two points of the best model in the world on graduate-level reasoning, so a Qwen3.8 leading that benchmark is entirely believable. Now look at SWE-bench Verified, where Fable 5's 95% sits nearly 15 points clear. Closing that in one generation would be extraordinary, and Kimi K3, Inkling, and MiniMax M3 all tried and fell short this month.
My prediction: Qwen3.8 lands top three on reasoning benchmarks, top five on coding, and first on price-to-performance. Second only to Fable 5 overall is a marketing sentence, not a measurement, until someone publishes a table.
We ran the previous generation through exactly this test in our Qwen3.7-Max vs Claude and Kimi Code Arena comparison, and the July 2026 model ranking tracks the verified field as it stands today.
The only comprehensive program designed to take you from basic prompting to building interactive Artifacts, custom integrations, and deploying production-ready code with Claude Code.
What to Watch Before You Migrate
Do not move production workloads to Qwen3.8 on the strength of a teaser. Five specific things need to land first, and each one is checkable in under a minute once it exists.
- An official Qwen blog post or model card with a published benchmark table, the same way Qwen3.7 and Qwen3.6 were documented.
- The active parameter count, not just the 2.4T total, since that determines serving cost and whether self-hosting is realistic.
- A Hugging Face repository with a real license file, if the open-weight promise is going to mean anything.
- Published API pricing, to confirm Alibaba is holding its value position rather than following Moonshot's premium pivot.
- Independent benchmark coverage from Artificial Analysis or similar, because launch numbers from any lab are marketing until a third party reproduces them.
Until those exist, the sensible move is to test Qwen3.8-Max-Preview through the official Alibaba console if you have access, benchmark it on your own workload, and keep your production traffic where it is. Your own evaluation on your own tasks beats every published number anyway, which is the one piece of advice in this post that will never go out of date.
Frequently Asked Questions
Q: What is Qwen3.8?
Qwen3.8 is Alibaba's next-generation flagship Qwen model, available now as Qwen3.8-Max-Preview through Alibaba Token Plan, Qoder, and QoderWork. The teaser claims 2.4 trillion parameters and a coming open-weight release. As of July 19, 2026 there is no official Qwen blog post, model card, or benchmark table, so treat those figures as unconfirmed claims.
Q: Is Qwen3.8 released yet?
Partially. Qwen3.8-Max-Preview is confirmed live and purchasable through Alibaba Token Plan subscription tiers, but the full model has not been formally launched with documentation. The latest Qwen flagship with a published benchmark table remains Qwen3.7-Max from May 19, 2026.
Q: How many parameters does Qwen3.8 have?
The announcement claims 2.4 trillion total parameters, which would make it the second largest publicly known model behind Kimi K3's 2.8 trillion. No official specification confirms this, and the more important number, active parameters per token, has not been disclosed at all.
Q: Will Qwen3.8 be open source?
The announcement says it will go open-weight soon, but this is unconfirmed and would break Alibaba's recent pattern. Both Qwen3.7-Max and Qwen3.6-Max-Preview shipped closed through Alibaba Cloud Model Studio. Alibaba does release smaller models like Qwen3-Coder-480B under Apache 2.0, so the claim is plausible but unproven.
Q: How can I try Qwen3.8-Max-Preview?
Access runs through Alibaba Token Plan, Qoder, and QoderWork, sold in Lite, Standard, and Pro subscription tiers rather than per-token pricing. The international Token Plan page is operated by Intelligent Cloud Computing (Singapore) Private Limited, so confirm that entity against your Alibaba Cloud account before entering payment details, and prefer the official Alibaba Cloud console where you have the choice.
Q: Is Qwen3.8 better than Kimi K3?
Unknown, because no Qwen3.8 benchmarks exist. For context, Kimi K3 has 2.8 trillion parameters and posts a verified 93.5% GPQA Diamond and 88.3% Terminal-Bench 2.1 at $3 input and $15 output. Qwen3.7-Max scores 92.4 GPQA at $1.25 input, so Qwen's edge has historically been price rather than peak capability.
Q: What is the difference between Qwen3.7 and Qwen3.8?
Qwen3.7-Max is verified and available now: 1M context, extended thinking, 92.4 GPQA Diamond, 80.4% SWE-bench Verified, $1.25 / $3.75, closed weights. Qwen3.8 is teased with roughly double the parameters and an open-weight release, but none of its specifications have been officially published yet.
Q: How much will Qwen3.8 cost?
No pricing has been published. For reference, Qwen3.7-Max costs $1.25 per million input tokens and $3.75 per million output tokens, and Alibaba has consistently priced its flagships well below Claude Fable 5's $10 / $50 and Kimi K3's $3 / $15. Expect Qwen to defend that value position.
Recommended Reads
- Qwen3.7-Max review
- Qwen3.7-Plus GUI agent review
- Qwen3.7-Max vs Claude Code Arena
- Qwen3.7 Max Preview arena ranks
- Qwen3.6-Max-Preview review
- Qwen3.6-27B beats 397B on coding
- Kimi K2.6 vs Qwen3.6 vs Opus 4.7
- Best AI models July 2026
We update this page the day Alibaba publishes official Qwen3.8 specs and benchmarks. Test every launch claim on your own workload before you trust it
