Qwen3.8-Max Preview: Alibaba’s 2.4T-Parameter Bid for the Frontier
Alibaba’s Qwen team announced Qwen3.8-Max-Preview on July 19, 2026 — a 2.4 trillion-parameter multimodal model the team calls its most capable system yet, claiming performance “second only to Fable 5” among the models it evaluated. The preview landed just two days after Moonshot AI’s 2.8 trillion-parameter Kimi K3 open-weight release, underscoring how quickly China’s frontier labs are now trading blows. Alibaba says Qwen 3.8 will go open-weight “soon,” though no date, license, or benchmark table has been published.
Intermediate
What Was Announced
Qwen3.8-Max-Preview is a sparse Mixture-of-Experts model with 2.4 trillion total parameters — the Qwen team’s first multimodal model to cross the 1 trillion-parameter mark. Qwen developer Shuai Bai highlighted exactly that milestone, calling it the team’s “first multimodal model above 1 trillion parameters.” The model processes text, images, video, and documents, and inherits the 1 million-token context window introduced with Qwen3.7-Max. The number of active parameters per token — the figure that actually determines serving cost in an MoE design — has not been disclosed.
Alibaba says the new model should outperform Qwen3.7-Max especially in coding and complex productivity tasks such as full-stack development, data analysis, and office workflows. For reference, Qwen3.7-Max (May 2026) posted 92.4 on GPQA Diamond, 80.4% on SWE-bench Verified, and 69.7 on Terminal-Bench 2.0 — so the bar the new model must clear is already near the frontier.
The Benchmark Gap
The boldest claim — that Qwen3.8 is “second only to Fable 5” — currently rests entirely on internal evaluation. As of the announcement, there is no published benchmark table, no model card, and no third-party scores on SWE-bench, GPQA, AIME, or LMSYS Arena. Until independent evaluations arrive, the near-frontier claim is unverifiable, and any figures circulating for “Qwen 3.8” are in practice Qwen3.7-Max numbers.
Availability and the Kimi K3 Context
The preview is live on Alibaba’s Token Plan, Qoder, and QoderWork at 10% of standard pricing during the preview period, with OpenAI- and Anthropic-compatible API protocols. Qwen3.7-Max pricing runs $1.25 per million input tokens and $3.75 per million output tokens, which gives a rough sense of where the new model may land.
The timing is hard to read as anything but competitive. Moonshot AI released Kimi K3 — a 2.8 trillion-parameter open-weight model — on July 17, and Alibaba’s announcement followed within 48 hours at the World AI Conference in Shanghai. Moonshot reportedly reached $300 million in annual recurring revenue in June and is planning an IPO within six months. By promising open weights for a 2.4T-parameter flagship, Alibaba is pressuring Moonshot’s strategy of keeping Kimi’s strongest access tiers behind its own app and API — and racing to claim the “best open model” title before anyone else does.
What This Means
For researchers and practitioners, two things are worth watching. First, whether the open-weight promise materializes with a usable license — a 2.4T-parameter checkpoint would be by far the largest openly released multimodal model, even if running it locally remains out of reach for all but large clusters. Second, whether independent benchmarks validate the frontier claim: the pattern of announcing capability ahead of evidence has become common in this race, and the community’s third-party evaluations will be the real test. Either way, the two-day cadence between trillion-parameter Chinese open-weight announcements signals that the open-model frontier is now moving at the same speed as the closed one.
Related Coverage
- Qwen3.6-27B: A Dense 27B Model That Beats a 397B MoE on Coding — the previous generation’s dense open-weight release
- Qwen3.6-35B-A3B: Alibaba Open-Sources a Frontier-Class Agentic Coder — the first open-weight Qwen3.6 model
- Qwen 3.5: Alibaba’s Native Multimodal Agent Model Arrives — the 397B flagship that started the natively multimodal Qwen line
- Junyang Lin Steps Down as Qwen Tech Lead in Abrupt Departure — leadership change earlier this year at the Qwen project
沪公网安备31011502017015号