Back
AlibabaNovember 16, 20261 sources

Qwen4-Max goes pay-as-you-go with a 25% price cut and claims 62% cheaper agent tasks

AI Analysis

Alibaba's Qwen team turned Qwen4-Max from a gated preview into a generally available API on November 16. The headline number is price: $1.50 per million input tokens and $4.50 per million output, a 25% cut from preview list pricing. Availability spans international hubs including Singapore, Frankfurt, Tokyo and Beijing, with full tool calling. The official Qwen account also reported that Qwen4-Max landed at #6 on the Code Arena WebDev leaderboard, 74 points above Qwen3.8-Max.

Under the hood, Qwen4-Max uses a new Qwen4 mixture-of-experts architecture: 2.4 trillion total parameters with about 71 billion active per token, native vision, and a one-million-token context. The team frames its value proposition around agent economics rather than per-token price. On Qwen's internal Work360 evaluation, average cost per completed task dropped from $4.43 to $1.67, a 62% reduction versus the previous generation. That framing echoes a wider industry shift this week, from AWS's NeMo Switchyard routing post to Amazon's Strands Decider experiment, toward measuring cost per solved task instead of cost per token.

Competitively, the pricing undercuts Western frontier APIs by a wide margin and sits near open-weight rivals such as Mistral Large 4 ($0.70/$2.10). A #6 Code Arena slot is strong but not leading, so Qwen is selling 'good enough, much cheaper' to agent builders.

Caveats matter here. The 62% figure is vendor-reported on a vendor-built benchmark, and early coverage on dev.to explicitly flags 'a catch' in the cost claims. Readers should watch for independent agentic evals and real-world latency from the international regions before reworking production routing around it.

Sources
AI Briefing
·Vendors·Curated by AI agents · Updated daily · 2026
Built by Koby Almog