What Qwen3.8 Is (2.4T Parameters)
Qwen3.8 is a large language model announced on July 19, 2026 by Qwen, the AI team at China's Alibaba. Qwen is the team that has released the "Qwen3" series and others as open weights (a form in which anyone can download the model's own data), and Qwen3.8 is its latest flagship.
Qwen3.8 at a glance (as announced)
The Announcement and Its Scale
The announcement came as a post on the official X account. At 2.4 trillion total parameters, it sits just behind the 2.8 trillion of Kimi K3, announced on July 16, 2026. If the weights are released as promised, Qwen3.8 becomes one of the largest open-weight models to date.
Qwen3.8 is launching and going open-weight soon!🌐 With a massive 2.4T parameters, this model is continuously evolving. We believe it’s one of the most powerful model available today, compatible to leading frontier AI models , second only to Fable 5. — from the July 19, 2026 announcement post (2.4T parameters, open-weight promise)
"Second Only to Fable 5" Is Self-Claimed
The official post describes the model as performing just behind Claude Fable 5, the top-tier model from Anthropic. What deserves attention is that no third-party benchmark supporting that positioning had been published at the time of the announcement. Reporting has pointed out the same gap in independent verification. In other words, at this stage "second only to Fable 5" is an assessment issued by the developer itself. With Kimi K3, external tallies appeared within days of the announcement. The safe approach with Qwen3.8 is the same: wait for outside results before judging.
Alibaba hasn't published independent third-party benchmarks — from the note on the absence of third-party benchmarks
How to Use It and What It Costs
Alongside the announcement, the preview version Qwen3.8-Max-Preview became available. The entry points and the rough cost are as follows.
How Qwen3.8 is offered (as announced; plan prices from reporting)
Only the Max-Preview Is Available
At present, the only way to try Qwen3.8 is the preview version bundled into Alibaba's "Token Plan" subscription and its coding-oriented services. Because the weights have not been published, you cannot yet download it and run it in your own environment. During the preview period it is reported to be available at 10% of the standard price.
A preview is available through Alibaba's Token Plan, Qoder, and QoderWork at 10 percent of the standard price. — from the description of preview availability and pricing
A Sense of the Plan Pricing ($6 to $68)
The Token Plan runs on credits in 7-day units. According to reports, the $6 Lite plan is allotted 2,500 credits and the $68 Pro plan 40,000 credits. The official pricing page likewise presents the plan as a cheaper way to reach Qwen3.8-Max-Preview. Details on how to sign up from outside China, and on payment in local currency, were not spelled out at the time of the announcement, so it is worth checking Alibaba's latest information before subscribing.
Get advanced models like Qwen3.8-Max-Preview for less — from the Token Plan description
Open-Weight Timing and the Kimi K3 Race
The most pressing question about this announcement is when the weights arrive. For now there is only the promise of "soon," with no fixed date.
Scale of large China-born open-weight models (total parameters)
"Soon," with No Date Committed
The phrasing on the official X account is "going open-weight soon," with no commitment on when. Checking the official Qwen page on Hugging Face (the site where AI models are distributed) on July 20, 2026, there was still no model card for Qwen3.8. That gap—a promise of "soon" without a committed date—is the clearest difference right now from Kimi K3, which has named July 27 as its deadline.
The Race with Kimi K3 and GLM-5.2
Qwen3.8 was announced three days after Kimi K3. Kimi K3 has 2.8 trillion total parameters and states plainly that its full weights will be released by July 27, 2026. Together with the already-released GLM-5.2 (744B), Chinese developers are running a tightly spaced race to put top-tier performance out as open weights. For how to choose and run open models in your own environment, see also practical coding with local LLMs.
The full model weights will be released by July 27, 2026. — from the note on Kimi K3's planned weights release
For overseas models like Qwen3.8, the primary sources start with official X posts and English blogs. If you want an AI to summarize a long English page, converting it to Markdown first preserves the structure of headings and tables and improves how accurately it is read.
Qwen3.8 leads with a scale of 2.4 trillion parameters and a confident "second only to Fable 5" positioning, yet neither the third-party benchmarks that would back it up nor a weights release date has arrived—it is an announcement-first model. If you want to try it today, the preview through the Token Plan is the only entry point, and running it locally means waiting for follow-up news on the weights. A realistic way to engage is to hold the performance claim in reserve until external verification appears, and to track Qwen3.8 alongside Kimi K3, whose release is proceeding on a stated date, through the end of July.
