AGENTIC COMMONSAI industry briefings

繁中EN

TagQwenPublished

All English articles

Qwen4-Max Launches: Native Vision and 1M Tokens per Request

Summary: Alibaba’s Qwen team announced Qwen4-Max on X on November 16, 2026: a flagship model built on a new mixture-of-experts architecture with native vision and a 1M-token context window. It is already listed in Alibaba Cloud Model Studio for complex reasoning, coding, and visual understanding, so teams on that platform can point an API call at it today rather than wait on a waitlist. (Earlier: Qwen4-Max Preview.)

The claims, and the one that got cut off

The announcement’s headline numbers: Qwen4-Max “outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over…” — and the post ends mid-sentence there. The direction of the cost claim is stated; the magnitude and the measurement method are not. Both numbers are the vendor’s own, and the sources contain no independent measurement.

Where it sits in the catalog

The Model Studio model list confirms the positioning: qwen4-max appears under both “Complex reasoning and coding” and “Visual understanding.” It sits alongside cheaper in-house tiers — qwen3.7-plus (“Balanced performance, speed, and cost”) and qwen3.8-flash (“Fast and cost-effective”) — and third-party text options including deepseek-v4-pro-0813, glm-5.3, kimi-k3, and MiniMax-M2.5.

Neither source states pricing, rate limits, regional availability, or data-retention terms. For a 1M-token context window, that pricing gap matters: long-context surcharges could offset a token-count saving, so the finished-task cost is currently unknowable from the announcement alone.

Our take: If you run long-document or image-heavy tasks on Qwen3.8-Max, spend an afternoon testing this — the model is live, the experiment is cheap and reversible, and a ~50% token cut would directly lower finished-task cost. But switch nothing yet: both efficiency numbers are vendor-measured, and the cost figure is literally unfinished. Assume long-context requests still carry surcharges until published pricing says otherwise. If your own run shows the token saving holding without quality regressions, switch; if qwen3.8-flash already clears your quality bar, this announcement changes nothing for you.

Sources

Original post
AGENTIC COMMONSOperated by PHLEGON LABS
ShareXEmail
Support us

Related reading

  1. Qwen3.5-Omni speech and multimodal API

    Multilingual voice builders can evaluate Qwen3.5-Omni; it supports recognition in 113 languages and dialects, with vendor benchmarks.

    Qwen

  2. Inkling 2 Small Goes Apache 2.0: Pull It or Host It

    Thinking Machines released Inkling 2 Small under Apache 2.0; Cloudflare Workers AI now hosts the same model.

    Open Models

  3. Qwen4-Max Preview Hits Alibaba Token with 1M Context

    Qwen teased Qwen4 on November 11, 2026, and Alibaba Cloud's docs now list qwen4-max-preview on the Token Plan with 1M context and 393,216 max output.

    Qwen