20 November 2026Explainers with cited sources
OmnistuNews
Tech & AI

Alibaba launches Qwen4-Max, a 2.4-trillion-parameter flagship priced far below OpenAI and Anthropic rivals

The mixture-of-experts model activates 71 billion parameters per token, handles a one-million-token context, and lists at $1.50/$4.50 per million tokens. Alibaba's claim of about 60% lower cost per task comes from its own benchmark and is not independently verified.

AI-assisted summary of linked sources · 20 November 2026 · How we report

Graphic: Alibaba launches Qwen4-Max, a 2.4-trillion-parameter flagship priced far below OpenAI and Anthropic rivals

Alibaba has launched Qwen4-Max, the first flagship of its new Qwen4 family, presenting it at the Qwen Conference Malaysia 2026 on 19 November, The Edge Malaysia reports; Alibaba announced the model on 16 November, when it went live on the QwenCloud API. The sparse mixture-of-experts model has 2.4 trillion total parameters with 71 billion activated per token, native vision, and a one-million-token context window.

The price pitch

Qwen4-Max lists at $1.50 per million input tokens and $4.50 per million output tokens on Alibaba Cloud's international tier, per the company's official pricing page - roughly a fifth of the $8/$40 list prices OpenAI and Anthropic publish for GPT-6.2 Astra and Claude Fable 5.2, and below the $2/$10 introductory rate Google lists for Gemini 4 Carbon through 30 November. Alibaba goes further: on its own Work360 benchmark covering 721 specialist fields, it says the model uses about 35% fewer tokens at high reasoning effort than its predecessor Qwen3.8-Max and that the average cost per finished task fell by about 60% - a claim no outside lab has verified.

Capability and safety claims

Alibaba says Qwen4-Max ranks sixth on Code Arena, a real-world coding leaderboard, and showed off two company-run demonstrations: a continuous 48-hour agent run for chip-cooling research and a 454-tool-call molecule-design workflow. On safety, the company says it filtered training data by intent - keeping defensive security material while removing offensive attack content - and that internal evaluations found no evidence the model significantly improves a user's ability to carry out dangerous tasks; Yahoo Tech reported on 19 November that the safety claims rest on Alibaba's own reviews and that no external red team had tested the model. Unlike smaller Qwen releases, the Max weights are closed; Alibaba says open-weight Qwen4 models will come later, with no date given.

Why it matters

A credible flagship at a fifth of rival list prices pressures the whole frontier tier's economics, and Alibaba is explicitly framing the competition as cost per finished task rather than benchmark rank. For enterprises, the savings case is attractive if the unverified efficiency claims hold up outside the vendor's own tests.

What to keep in mind

Almost everything quantitative here is Alibaba's own reporting: the Work360 savings, the Code Arena rank, the safety evaluations and the research demos are vendor-run and not independently reproduced. List prices are verifiable, but real-world cost per task depends on workload.

More from Omnistu News