Qwen3.8-Max goes official on QwenCloud; Max-class open weights next week

Qwen3.8-Max is now an official product post, not only a July preview teaser. Hosted access is live on QwenCloud. Max-class open weights are pegged to **next week**, not a public weight dump today. The 2 August 2026 blog calls it the most capable model in the Qwen family so far, scales it to **2.4 trillion parameters with 95B active** on the Qwen 3.5 architectural foundation, and frames the release around long-horizon coding and real-world cowork.

From the same window, @Alibaba_Qwen on X (3 August 2026 UTC) said:

"Meet Qwen3.8-Max - our most capable model to date. Next week, the open weights of Qwen3.8-Max will be released, and Qwen3.8-27B is also going open-weights to meet you all! Qwen3.8-Max, a new bar for coding and cowork at 2.4T parameters"

The blog's integration snippet names API id `qwen3.8-max` with text and image input, a 1M-token context window, and 65,536 max tokens. Try surfaces also include Qwen Studio / Chat.

Key facts

• Official claim: first open-source of a **Qwen-Max-class** weight set; timing remains **next week** (watch Hub and ModelScope when files appear).
• Hosted now via QwenCloud / compatible API path; blog demos stress multi-day coding and profession workflows.
• Vendor coding showcase: autonomous run on public qwen-code-dev-bot/oh-my-cli; blog cites about 16 days through 30 July 2026 with 265 commits, 127 PRs, and 151 issues (vendor-run).
• Vendor contest showcase: Tianchi WWW2025 multimodal intent task; blog cites a 24-hour autonomous run at 0.853 accuracy, ahead of 458 of 526 human teams (vendor report).
• X also plans **Qwen3.8-27B** open weights in the same week; treat 27B as a separate smaller drop until a model card lands.

How to read it

Blog plus Qwen X are the primaries. Hosted availability and the 2.4T / 95B active line are on-record from Qwen. Coding stats, contest rank, and profession demos stay vendor evaluations until independent harnesses and a public license file appear. "Next week" is a calendar hedge. For self-hosting, wait for the model card, shards, and license rather than assuming every hosted behavior ships in the open checkpoint.

Independence

Frontier Flux is not affiliated with Alibaba or the Qwen team. Social posts are alerts; the blog is the product primary. Vendor demos and rank language are not our rankings.