Alibaba's Qwen3.8-Max-Preview: 2.4 Trillion Parameters, Claimed Second Only to Fable 5

On July 19, 2026, Alibaba's Qwen team announced the Qwen3.8-Max-Preview — a 2.4 trillion-parameter flagship AI model — alongside a commitment to release open-weight versions of Qwen3.8 in the future. Here is a breakdown of every key claim with cited sources.

Parameter Count: 2.4 Trillion

The model is confirmed at 2.4 trillion (2.4T) parameters across all major reports. Alibaba's official X account stated the model has "a massive 2.4T parameters" , Bloomberg (via Yahoo Finance) reported the same figure , and Chinese outlets like IT之家 and Sina Finance all confirmed 2.4T .

Claimed Performance Ranking: "Second Only to Fable 5"

Alibaba is publicly positioning Qwen3.8 as "second only to Anthropic's Fable 5" among all models it benchmarked against. The official Qwen X account stated the model is "compatible to leading frontier AI models, second only to Fable 5" . Bloomberg reported the same language: Alibaba described Qwen3.8 as "comparable to leading frontier AI models and second only to Anthropic PBC's Fable 5" . Chinese media quoted the phrasing "可能是除了 Fable 5 外最强大的模型" ("possibly the most powerful model besides Fable 5") .

Availability: Qwen3.8-Max-Preview to Developers

The preview version launched on July 19, 2026 on three Alibaba platforms:

• Token Plan — Alibaba's new subscription-based AI credits system. Personal plans start at ¥35-39/month with a launch discount: daytime credits at 10% cost, and nighttime (22:00–08:00) at an additional 80% off (effectively 0.2% of normal rate) .
• Qoder — Alibaba's coding platform .
• QoderWork — Alibaba's developer work platform .

The model also became available on Qwen's PC and Web clients on the same day . Access is available to developers in mainland China, though Bloomberg's report indicates broader availability was being rolled out .

Planned Open-Weight Release (No Date Given)

Alibaba stated that the full Qwen3.8 model will be released open-weight "soon," with the official X account posting "Qwen3.8 is launching and going open-weight soon!" . Chinese reports confirm "正式版将于近期发布并开源" ("the official version will be released and open-sourced soon") . This marks a notable shift: Alibaba had kept its recent Max-series models (Qwen3.7-Max, Qwen3.6-Max) proprietary, whereas Qwen3.8 returns to the open-weight approach seen with earlier models like Qwen2.5 and Qwen3 base .

No Independent Third-Party Verification

As of launch day, no independent benchmarks have been published for Qwen3.8-Max-Preview. The BuildFastWithAI analysis noted that "Alibaba has published exactly zero benchmarks for it" — the product pages mention the model in only a single promotional line . All performance claims are based on Alibaba's internal evaluations across tasks like full-stack development, data analysis, and Office workflows . This mirrors the pattern seen with earlier Qwen Max models (Qwen3.5, Qwen3.7) where self-reported benchmarks were released without external audit . There is no LMArena or third-party leaderboard result yet for this specific model.

Broader Context: The 2026 US-China AI Race

The Qwen3.8 launch is the latest salvo in an increasingly aggressive 2026 AI competition between Chinese and American labs . Key dynamics:

• Scale race: Qwen3.8's 2.4T parameters makes it one of the largest models publicly previewed by any Chinese lab, following a rapid cadence of releases (Qwen3.6-Max in April 2026, Qwen3.7-Max in May 2026, now Qwen3.8 in July 2026) .
• Catch-up with frontier US labs: Alibaba explicitly benchmarks against Anthropic's Fable 5 (the current frontier US model), claiming second place. Earlier Qwen3.7-Max placed around 13th globally on LMArena text benchmarks .
• Open-weight strategy as a differentiator: While US frontier labs (OpenAI, Anthropic, Google) keep their most capable models proprietary, Alibaba's pledge to open-weight Qwen3.8 is a strategic move to attract developer mindshare and accelerate ecosystem growth in China and globally .
• Geopolitical stakes: Chinese AI labs are under pressure to demonstrate parity with US frontier models amid export controls on advanced chips. Alibaba's rapid release cycle and large-scale parameter counts signal that Chinese labs are investing heavily to close the gap .

Bottom line: The Qwen3.8 announcement is heavy on ambition and scale but light on verified evidence. All claims of "second only to Fable 5" remain unsubstantiated by independent testing, and the absence of published benchmarks invites caution until third-party evaluations emerge.