# Qwen3.8: a 2.4-trillion claim that arrived before the proof

> Alibaba previewed a 2.4T-parameter model it says is second only to Fable 5 — and shipped that claim with no benchmarks, no model card, and no independent testing.

_Source: Alibaba / Qwen announcement on X (vendor, self-reported) · 2026-07-21 · 5 min read_

Canonical: https://iyu.app/e/qwen3-8-max-preview-2p4t

## The 60-second version

Alibaba's Qwen team previewed Qwen3.8, a 2.4-trillion-parameter sparse-MoE multimodal model going open-weight "soon," and claimed it is "second only to Fable 5" — with no benchmarks, model card, or independent testing to back it.

**Key points**

- Concrete: 2.4T total params (sparse MoE), multimodal, and a live Max preview via Token Plan / Qoder / QoderWork.
- Unverified: the "#2 in the world" ranking is a vendor claim with zero published evidence.
- Undisclosed: active params, context window, benchmarks, license, and the actual open-weight date.
- Context: arrives days after Kimi K3 — a real Chinese open-weight surge, which is the durable story.

**Verdict.** The trend (fast, open-weight, frontier-scale models from China) is real and worth tracking; the specific 'second only to Fable 5' claim is marketing until an independent eval — and a license and date — show up.

## Full explainer

> **⚑ Caveat:** Everything here is **Alibaba's own announcement**. The "second only to Fable 5" ranking shipped with **no benchmarks, no model card, and no independent testing** — treat it as a vendor claim, not a measured result.


### What was announced — A 2.4-trillion-parameter preview

Alibaba's **Qwen** team said its next model, **Qwen3.8**, is launching and going **open-weight soon**, and put a preview — **Qwen3.8-Max-Preview** — in front of developers right away through Alibaba's Token Plan subscriptions and its coding tools **Qoder** and **QoderWork**. So you can already try it; you just can't yet verify the boast attached to it.

The concrete details: **2.4 trillion total parameters** in a **sparse mixture-of-experts** design (only a slice of those parameters activate per token, which is how a model this big stays runnable), described as Qwen's first model above a trillion parameters to go **multimodal** (text + images). Full open weights are promised, but not dated.

- **2.4T** — total parameters (sparse MoE; vendor figure)
- **"#2"** — self-claimed rank — "second only to Fable 5"
- **0** — benchmarks / model cards published


### The claim — "Second only to Fable 5"

Qwen framed the model as one of the most powerful available today, compatible with leading frontier models and **"second only to Fable 5"** — a claim to the number-two slot in the entire field. That's a striking statement. It's also, right now, unaccompanied by anything you could use to check it.

> The preview arrived before the proof.


### What's missing — A lot of blanks for a #2 claim

For a model claiming second place in the world, the announcement is notably light on specifics. Independent write-ups flagged the same gaps: no scores, no model card, no license, no firm open-weight date.

- **:** Stated
- **Size:** 2.4T total params, sparse MoE
- **Modality:** Multimodal (text + images)
- **Ranking:** "Second only to Fable 5"
- **Openness:** Open weights "soon"
- **Training:** —


### Why it matters — The pattern, not the boast

The ranking is marketing until someone outside Alibaba runs the numbers — but the **trend** is real. Qwen3.8 lands just days after Moonshot's [Kimi K3](/e/kimi-k3-open-3t-model), another giant Chinese open-weight model, and feeds directly into the live debate over [whether Chinese labs have closed the gap](/e/china-ai-frontier-gap) to the US frontier. Chinese labs shipping frontier-scale models **and promising open weights** is a genuinely different competitive posture from the closed US frontier. That momentum isn't hype; the specific #2 claim, for now, is.

> **🔎** **What would change the picture:** published benchmarks or an independent eval, a model card with the active-parameter count and context window, and — most of all — the open-weight **license and release date**. Those decide whether this is usable or just impressive.


### The bottom line — Big if true

- Qwen3.8: 2.4T-param sparse-MoE multimodal model, open weights promised "soon" (vendor figures).
- A Max preview is already testable via Alibaba's Token Plan, Qoder, and QoderWork.
- The "second only to Fable 5" ranking ships with no benchmarks, model card, or independent test.
- Lands days after Kimi K3 — part of a real China open-weight surge, separate from this one claim.
- Watch for the license and open-weight date; that's what decides if it matters to you.


## Primary sources

- [Qwen (@Alibaba_Qwen) — Qwen3.8 launch post on X](https://x.com/Alibaba_Qwen/status/2078759124914098291)
- [MarkTechPost — Alibaba previews Qwen3.8-Max (2.4T)](https://www.marktechpost.com/2026/07/19/alibaba-previews-qwen3-8-max-a-2-4-trillion-parameter-multimodal-model-days-after-moonshots-kimi-k3-open-weight-launch/)

---
_Published by iyu (https://iyu.app) — the day's AI news, checked against primary sources and rewritten in plain language. Free to quote with attribution and a link to the canonical URL._