Breaking

Alibaba Unveils Qwen 3.8 Max Preview as China Doubles Down on Open-Weight AI Strategy

July 20, 2026 at 06:18 EDT

Foundation Models
Open Source
Software Dev & Coding

Alibaba on July 19, 2026 released a preview of its new flagship large language model, Qwen 3.8 Max, a 2.4-trillion-parameter system that the company says rivals leading frontier models, capping a week that underscored China's distinctive bet on open-weight AI.

July 19, 2026 · China's AI Race

China Doubles Down on Open-Weight AI With Trillion-Parameter Giants

Alibaba's 2.4-trillion-parameter Qwen 3.8 Max lands days after Moonshot's 2.8T Kimi K3 — twin bets that publishing model weights, not locking them away, is the path to a global developer ecosystem.

2.4T

Qwen 3.8 Max parameters

2.8T

Kimi K3 parameters

Kimi K3 context-window tokens

Two flagships, head to head — parameters (trillions)

2.4T

Qwen 3.8 Max

Alibaba

2.8T

Kimi K3

Moonshot AI

Open weights on the way

Unlike U.S. rivals guarding proprietary models, both firms plan to publish their weights — a distinctive bet made under semiconductor export limits.

Qwen 3.8 Max

Open-weight release: coming days

Kimi K3

Open-weight release: ~July 27

Kimi K3 API pricing — $3 per M input tokens · $15 per M output tokens

Momentum

Kimi K3 tops the Frontend Code Arena
First open model to beat proprietary systems on a full web-engineering benchmark
Low-cost access builds developer mindshare

Caution

Benchmarks aren't the whole picture
Little public data on speed, hallucination, fine-tuning
Surging demand forced Moonshot to pause sign-ups on strained GPUs

The coming weeks test whether open-weight momentum converts into durable adoption against well-funded proprietary rivals — as independent evaluations remain underway.

Continue reading

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