Announced Three Days After Kimi K3
Alibaba announced Qwen 3.8-Max on Sunday, July 19, 2026, via an official Qwen post on X, timed to coincide with WAIC Shanghai — three days after Moonshot's Kimi K3 launched. The timing is hard to read as coincidence: it is the second large Chinese open-weight-class model announcement inside a week, following a run that also includes GLM-5.2 and DeepSeek V4.
What's Confirmed
Qwen 3.8-Max is a sparse mixture-of-experts model at 2.4 trillion total parameters — Alibaba has not disclosed the active-parameter count. It is the first Qwen model over 1 trillion parameters that is multimodal, with Alibaba claiming support for text, images, video, and documents. It is preview-only: available through Alibaba's Token Plan subscription and the Qoder / QoderWork coding platforms. As of this writing it is not documented on Alibaba Cloud Model Studio, not listed on OpenRouter, and has no confirmed public model ID, pricing, or context window.
What's Marketing, Not Benchmark
Alibaba's announcement describes Qwen 3.8-Max as “second only to Claude Fable 5” — a strong claim with no benchmark numbers published to back it up. Multiple outlets covering the announcement flagged the same gap: no Artificial Analysis Intelligence Index score, no SWE-bench or Terminal-Bench figures, nothing independently verifiable yet. Compare that to Kimi K3's launch, which shipped without a technical report but at least has third-party Artificial Analysis numbers available now (see our K3 benchmarks post). Until Qwen 3.8-Max has equivalent independent evaluation, “second only to Fable 5” is a claim to note, not a number to plan around.
Open Weight “Soon”
Alibaba says Qwen 3.8-Max is “going open-weight soon,” with no date and no license specified — less concrete than Moonshot's committed July 27 date for K3's weights (see our Kimi K3 open-source post). Alibaba's stock rose roughly 3% pre-market on the announcement, suggesting the market is pricing in Qwen staying competitive in the frontier race regardless of how thin the benchmark disclosure is today.
Where This Fits
Qwen 3.8-Max is best read right now as a marker in the July 2026 China open-weight race rather than a model you can evaluate yet. Alibaba's existing Qwen 3.7 is already usable today if you want a current Alibaba model in production. Once 3.8-Max reaches a public API or shows up on OpenRouter, it will slot into the same routing setup as every other OpenRouter model and become something you can point OpenClaw or Hermes Agent at directly.
Running Models on an Always-On Agent
Whichever model ends up ahead on real benchmarks — Fable 5, K3, or eventually Qwen 3.8-Max — the point of OpenClaw Launch is that switching between them does not mean rebuilding anything. The model is a dropdown, you bring your own key, and your agent keeps running on Telegram, Discord, WhatsApp, WeChat, or the web while the underlying model changes underneath it. See the full model list for what is supported today.