Alibaba Previews Qwen3.8-Max, a 2.4T Model It Says Is Second Only to Fable 5

TLDR

Alibaba’s Qwen team on July 19, 2026 announced Qwen3.8 and made Qwen3.8-Max-Preview available the same day on Alibaba’s Token Plan, Qoder, and QoderWork. The company describes a 2.4-trillion-parameter multimodal model “comparable to leading frontier AI models, second only to Fable 5,” with open weights promised soon. Launch-week promotions on Qoder cut Credits consumption for the preview tier by 90% (and up to 98% off-peak SGT). Independent Arena / Artificial Analysis ranks were not yet published at announcement.

What shipped

From the official @Alibaba_Qwen announcement (Jul 19), Qoder launch docs, SCMP, and THE DECODER same-day coverage:

Item	Detail
Model	Qwen3.8 family; flagship tier Qwen3.8-Max-Preview
Scale	2.4T parameters (company claim)
Modality	Multimodal — team notes image/video/document inputs; first Qwen model above 1T with full multimodal framing (per researcher posts)
Access (day one)	Token Plan (Alibaba Cloud / international seats), Qoder, QoderWork
Open weights	Promised soon — no date, license, or exact Max-checkpoint guarantee in the launch post
vs prior	Positioned above Qwen3.7-Max, especially coding and long-horizon “Cowork” productivity
Promo	Qoder: 0.05× Credits vs 0.5× standard for Max-Preview (90% off); night window 0.01× (98% off), SGT

The drop landed mid-WAIC 2026 in Shanghai (Jul 17–20) and three days after Moonshot’s Kimi K3 (2.8T, open-class path). Market reads framed Qwen’s preview + open-weight teaser as a direct answer to K3’s momentum—though K3 already posted independent Arena/AA scores while Qwen’s launch table was still empty.

Independent rankings

Mandatory check, launch day: public LMArena / Chatbot Arena and Artificial Analysis listings for Qwen3.8-Max-Preview were not available in the Jul 19–20 research window (model too new / not yet boarded). Treat Alibaba’s “second only to Fable 5” line as vendor claim only until third-party Elo or AA Intelligence Index numbers appear. Do not invent ranks.

Price and access shape

International Team Token Plan seats remain the $30 / $100 / $200 monthly structure (token/caching/tool-use dependent). Qoder’s launch campaign is Credits multiplier, not a permanent API list-price cut—and Experts-mode / some sub-agent calls stay at regular rates. Full open-weight shipping details (active params, context length, quantization, license) were not in the first-day materials.

Why this story matters

China’s open/near-open frontier track now has two multi-trillion-parameter flags in the same week: Kimi K3 (Moonshot) and Qwen3.8 (Alibaba). Even without independent boards, a live 2.4T Max preview on coding surfaces forces Western closed labs and enterprise buyers to re-price “wait for open Chinese weights” as a procurement timeline, not a research footnote—especially while WAIC sells Beijing’s open-cooperation narrative on the global stage.

Sources

• Qwen (@Alibaba_Qwen) primary announcement (July 19, 2026). https://x.com/Alibaba_Qwen/status/2078759124914098291
• Qoder docs: Qwen3.8-Max-Preview launch / Credits campaign (start July 19, 2026). https://docs.qoder.com/events/qwen-max-preview
• SCMP: “Alibaba says newest Qwen AI model is second only to Anthropic’s Claude Fable 5” (July 19, 2026). https://www.scmp.com/tech/article/3361119/alibaba-says-newest-qwen-ai-model-second-only-anthropics-claude-fable-5
• THE DECODER: “Alibaba’s Qwen takes on Kimi K3 with open-weight Qwen 3.8…” (July 19, 2026). https://the-decoder.com/alibabas-qwen-takes-on-kimi-k3-with-open-weight-qwen-3-8-says-model-is-second-only-to-fable-5/

Featured Image Alt Text

Engraved Alibaba Qwen mark with 2.4T scale rings and open-weight calendar cue for the July 19 Qwen3.8-Max preview.

Tags

Alibaba, Qwen, Qwen3.8, Models, China, Open Weights, Qoder, Token Plan, Frontier, July 19