Alibaba's Qwen team announced Qwen3.8 on Sunday, July 19, 2026, a 2.4-trillion-parameter frontier model the company says will be released with open weights. Alibaba claims the model is "one of the most powerful models available today," positioning it as second only to Anthropic's Fable 5 — and a preview version, Qwen3.8-Max-Preview, is already available for testing on Alibaba's Token Plan, Qoder, and QoderWork platforms.
Key Highlights
- Qwen3.8 packs 2.4 trillion parameters, making it one of the largest models Alibaba has ever built
- Alibaba has committed to an open-weight release, though no date has been announced yet
- Qwen3.8-Max-Preview is live today on Token Plan, Qoder, and QoderWork
- Alibaba positions the model as second only to Anthropic's Fable 5 among frontier AI systems
- No independent benchmarks have been published yet; the claim rests on Alibaba's internal evaluations
Details
The announcement came directly from the official Qwen account on X, describing a model that is "continuously evolving" at 2.4 trillion parameters. Rather than waiting for the full open-weight release, Alibaba made the preview immediately accessible through its consumer and developer channels.
The distribution strategy is notable. Qwen3.8-Max-Preview debuts inside Alibaba's Token Plan, a recently unified subscription that pools text, image, and video generation quotas into a single balance, priced at roughly 4 dollars per month after coupons. The same plan bundles competing Chinese models, including GLM-5.2 and DeepSeek-V4-Pro, alongside Qwen — a sign that Alibaba is competing on distribution and price as much as raw capability.
Early testers report the preview is functional but not yet fast: one developer measured throughput fluctuating between 22 and 55 tokens per second, well below the responsiveness of smaller models like GLM-5.2.
Impact
Qwen3.8 lands just two days after Moonshot AI announced Kimi K3, a 2.8-trillion-parameter model billed as the largest open model ever, with weights scheduled for release on July 27. Within a single week, two Chinese labs have promised frontier-class open-weight models in the multi-trillion-parameter range — a pace that is compressing the gap between open and closed AI from months to weeks.
For developers and businesses in the MENA region and beyond, the implications are mostly about cost. Frontier-class open weights mean cheaper inference through third-party hosts, more room for self-hosting sensitive workloads, and stronger price pressure on closed-model APIs.
The community reaction, however, highlighted a tension: a 2.4-trillion-parameter model is far too large for local deployment. Replies to the announcement, including from the quantization specialists at Unsloth AI, immediately asked for smaller and medium-sized variants that can actually run on accessible hardware.
Background
Qwen3.8 follows Qwen3.7-Max, which Alibaba presented as the anchor of its "full-stack AI factory" strategy, and a string of increasingly aggressive open releases from Chinese labs in 2026, including Zhipu's GLM-5.2, Ant Group's trillion-parameter Ling 2.6, and Moonshot's Kimi series. Alibaba's Qwen family has consistently ranked among the most downloaded open-weight model lines in the world.
What's Next
Alibaba has not given a date for the open-weight release or published benchmark results, both of which the community is actively demanding. Watch for the full Qwen3.8 launch, official evaluations against Fable 5 and Kimi K3, and whether Alibaba answers the loudest request in the replies: smaller distilled variants that developers can run themselves.
Source: Alibaba Qwen