• Tech Support ⤴
  • Projects
  • Services
    • AI Development
    • UI/UX Design
    • Web Development
    • Technology Support
    • Mobile App Development
    • Banking ATM Interfaces
    • Process Automation
    • Security Auditing
    • Local AI Servers
  • odoo ERP
get in touchStart with Eva
logo
Tech Support ⤴
Projects
Services
AI DevelopmentUI/UX DesignWeb DevelopmentTechnology SupportMobile App DevelopmentBanking ATM InterfacesProcess AutomationSecurity AuditingLocal AI Servers
odoo ERP
get in touchStart with Eva
Loading…
logo

Transforming businesses through AI-powered digital innovation and creative excellence.

Quick Links

BlogAinexProjectsContact us

Contact Us

pinDubai Digital Park, A5, DTEC - Silicon Oasisemail[email protected]phone+971 55 7538087
© 2026 aratech. All rights reserved.
Privacy PolicyTerms of ServiceCookie Policy
Home / Blog / Alibaba's Qwen3.8-Max Is Here With 2.4 Trillion Parameters. But Can You Actually Use It?

Alibaba's Qwen3.8-Max Is Here With 2.4 Trillion Parameters. But Can You Actually Use It?

Alibaba previews Qwen3.8-Max (2.4T params) claiming second only to Fable 5. But with no open weights, no benchmarks, and a pattern of going Max-first — is this a real release or a positioning play?

July 20, 2026 - 8 min read
Alibaba's Qwen3.8-Max Is Here With 2.4 Trillion Parameters. But Can You Actually Use It?

July 19, 2026 — WAIC Shanghai. Two days after Moonshot AI dropped Kimi K3 (2.8 trillion parameters, open weights promised for July 27), Alibaba fired back. The Qwen team previewed Qwen3.8-Max-Preview, a 2.4 trillion-parameter multimodal model that the company claims is "second only to Claude Fable 5" among the systems it benchmarked.

The AI arms race just got a lot louder.

But here's the thing — and it's a big one — the model you can actually use today is not the model you've been reading about. Let's separate what shipped from what was claimed.

What's Actually Available Right Now

The live product is Qwen3.8-Max-Preview, accessible through Alibaba's Token Plan subscription, Qoder, and QoderWork platforms. Shuai Bai, a Qwen developer, confirmed it's the team's first multimodal model above 1 trillion parameters — handling text, images, video, and documents in a sparse Mixture-of-Experts architecture.

Key facts about what's live:

  • 2.4 trillion parameters (Alibaba's claim — no model card or specification has been published to confirm this)
  • Sparse MoE architecture — the active parameter count per token is undisclosed, which is the number that actually determines serving cost
  • Multimodal — processes text, images, video, and documents
  • OpenAI and Anthropic API compatible — existing coding agents can point at Qwen3.8 without rebuilding their harnesses
  • Preview pricing — offered at 10% of standard pricing through Token Plan

The Open-Weight Problem

Qwen3.8 is being marketed as an open model. But today, there are no open weights. Not on Hugging Face, not anywhere.

The official @Alibaba_Qwen account says open weights are coming "soon" — but with no date, no license, no Hugging Face repository, and critically, no small variant announced. A 2.4 trillion-parameter model is a datacenter object no matter how hard you quantize it. Even at 1-bit quantization, you're looking at roughly 270GB+. For context, that's about 35 NVIDIA H200 GPUs just for the weights.

Alibaba's own history tells a mixed story here:

GenerationLaunch DateOpen Weights ArrivedConsumer-Sized Variant
Qwen 3.5Feb 16, 2026Feb 24 (8 days)9B (fits 8GB card)
Qwen 3.6April 2026Same day27B dense, 35B-A3B MoE
Qwen 3.7May 2026NeverClosed API only
Qwen 3.8July 19, 2026"Soon" (TBD)Not announced

The pattern shifted with Qwen 3.7, which shipped as a proprietary API-only model with no open weights — not a 27B, not a 9B, nothing. If Qwen 3.8 follows the 3.5/3.6 cadence, expect a runnable 27B or 35B-A3B variant within one to two weeks. If it follows the 3.7 precedent, you get a datacenter drop labeled "open" and nothing for your GPU.

"Second Only to Fable 5" — Let's Talk About That

Alibaba's "second only to Fable 5" claim is a self-reported ranking. No benchmark table has been published. No prompts, harnesses, or methodology. The Artificial Analysis leaderboard — the independent index most widely cited — doesn't list Qwen 3.8 at all. Its current top looks like this:

  1. Claude Fable 5 — 60
  2. GPT-5.6 — 59
  3. Kimi K3 — 57
  4. Claude Opus 4.8 — 56
  5. Qwen 3.7 Max — 46

There's historical precedent here: Qwen 3.7 Max was billed as "China's number one" and when independently scored, landed at 46 — still strong, just not what the announcement implied.

How It Stacks Up: The TrilogyAI Test

One independent benchmark has emerged. TrilogyAI ran Qwen3.8-Max-Preview and Kimi K3 through a matched StackPerf test — 269 files across two unfamiliar projects requiring planning, storyboarding, and system architecture recommendations.

Results after blind review:

MetricKimi K3Qwen3.8-Max-Preview
Score (out of 100)8380
Task completionSooner, fewer tokensLonger report, better metadata
Tool callsMore requestsFewer, 100% success rate
Cache hit rate>90%>90%

Kimi edged Qwen by 3 points, but Qwen produced stronger replay metadata and cleaner system boundary definitions. The researchers noted that "their shared recommendation was stronger than either proposed contract on its own."

The Geopolitical Context

This release doesn't exist in a vacuum. On July 7, Reuters reported that China's Ministry of Commerce held talks with Alibaba, ByteDance, and Z.ai about potentially restricting overseas access to China's most advanced AI models — talks that explicitly covered open-weight releases and even unreleased future models.

Alibaba was a named participant.

The company hasn't connected these talks to Qwen's release plans, and no rules exist yet. But the timing of "open weights soon" alongside government consultations about restricting them is worth watching closely.

The Bottom Line

Qwen3.8-Max-Preview is a real, capable model. If Alibaba's 2.4T claim holds up under independent testing, it's genuinely competitive with frontier systems. The multimodal capabilities, API compatibility, and preview pricing are all positives.

But until open weights ship — with a license that means something, in a size that fits real hardware — calling Qwen 3.8 "open" is premature. Watch the Qwen Hugging Face org, not the tweets. Weights or a named small variant is the only signal that counts.

For now, the AI arms race is real. The open model revolution? Still waiting.


This article was researched and published on July 20, 2026. Qwen3.8-Max-Preview was announced July 19, 2026 at WAIC Shanghai. Benchmarks and availability change rapidly — check the latest from @Alibaba_Qwen for updates.

Table of Contents

  • ↗What's Actually Available Right Now
  • ↗The Open-Weight Problem
  • ↗"Second Only to Fable 5" — Let's Talk About That
  • ↗How It Stacks Up: The TrilogyAI Test
  • ↗The Geopolitical Context
  • ↗The Bottom Line

Related Posts

GLM-5.2: The Open Model Nobody Can Ban

GLM-5.2: The Open Model Nobody Can Ban

When the US restricted Claude Fable 5, Zhipu AI released GLM-5.2 under MIT license. The open-weights era is no longer theoretical.

Necolas HamwiNecolas Hamwi
July 19, 2026 - 12 min read
Featured image showing 35 self-hosted GitHub projects highlighted in the article

35 Self-hosted Projects on Github: TaskView, ConvertX, Work-Review, relaticle, postlab, rejourney

We rounded up 35 standout self-hosted projects on GitHub. Here are the six that deserve your attention right now: TaskView, ConvertX, Work-Review, relaticle, postlab, and rejourney.

Necolas HamwiNecolas Hamwi
July 12, 2026 - 8 min read
35 Self-Hosted Projects on GitHub — Episode 5

35 Self-Hosted Projects on GitHub — Episode 5

From task management and file conversion to AI-powered job hunting and wildlife sound analysis — here are 10 standout self-hosted projects from GitHub that you can run on your own infrastructure.

Necolas HamwiNecolas Hamwi
July 12, 2026 - 0 min read