Breaking

Alibaba Launches Qwen3.8-Max-Preview, a 2.4-Trillion-Parameter Frontier Model, on Cloud and Chat

July 19, 2026 at 09:17 EDT

Foundation Models
AI Agents

Alibaba has begun offering Qwen3.8-Max-Preview, a 2.4-trillion-parameter model that its Qwen team claims ranks among the world's strongest, available immediately for testing across Alibaba Cloud and Qwen Chat. The preview surfaced on July 19, 2026, positioning the company for another round in an intensifying global race to ship ever-larger frontier systems.

July 19, 2026 · Alibaba Qwen

Qwen3.8-Max-Preview: a 2.4-Trillion-Parameter Bid for the Frontier

Alibaba has opened testing on the flagship of its Qwen3 line, claiming it ranks among the world's strongest models — available now via Alibaba Cloud and Qwen Chat, with a full open-weight release promised soon.

2.4T

Total parameters disclosed (active count kept private)

~400

Task agentic suite cited in leaked figures

Where Qwen frames it — second only to "Fable 5"

Unverified benchmark claims circulating at launch

Scale of parameter growth (top) and the two disputed score gaps (bottom). No formal benchmarks published — treat cautiously.

Parameter scale vs. predecessor

~1T

Qwen3-Max
(2025)

2.4T

Qwen3.8-Max
(preview)

Claimed score margins (points)

+5.0

Ahead of "Opus 4.8 Max" on the agentic suite

−12.6

Behind "Fable 5" on coding

Optimists say

Generous token consumption on the Token Plan
Clear gains on agentic workloads
Promising cost-performance

Skeptics say

Lagged Fable 5, Opus 4.8 Max & GPT-5.6 in practice
Weaker on longer tasks
Gap between marketing and observed capability

What decides the verdict

Whether Qwen3.8 lives up to its billing hinges on two things still missing.

Open-weight release

Full Qwen3.8 promised as downloadable, fine-tunable weights

Public benchmarks

No formal results published; independent testing still thin

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.