Qwen3.5 397B A17B API Prices Rise 54%: What It Costs You Now

Qwen has repriced its largest publicly available mixture-of-experts model, Qwen3.5 397B A17B, with increases hitting both sides of the token ledger.

The New Numbers

• Input: $0.39 → $0.50 per 1M tokens
• Output: $2.34 → $3.60 per 1M tokens
• Overall change: +54%

The output price hike is the sharper hit. At $3.60 per 1M output tokens, generation-heavy workloads — chatbots, long-form summarisation, agentic pipelines that produce verbose tool calls — will feel this more than retrieval or classification tasks that stay input-heavy.

Does It Matter for Your Workload?

Run a quick check: if your app generates more tokens than it consumes (output-to-input ratio above 1:1), your effective cost increase will be steeper than 54% on a blended basis. A pipeline burning 1M input and 2M output tokens monthly was costing roughly $5.07; the same run now lands at $7.70 — a $2.63 monthly jump per million input tokens at that ratio, before any volume scaling.

For lighter workloads or retrieval-augmented generation where output is brief, the absolute dollar impact stays manageable. For high-throughput inference, it's worth benchmarking alternatives.

What to Do Next

Check the Qwen3.5 397B A17B — live specs & price history page for up-to-date pricing and to compare against other large MoE models available through the site. If this increase pushes you over budget, the comparison filters let you sort by output token cost directly.

Price changes on frontier-class models aren't unusual as providers move away from introductory rates, but a 54% jump in one move is worth factoring into any contract or budget renewal coming up.

Ezra, Scout AI Team