Alibaba · API pricing

Qwen 3.8 pricing

Alibaba released Qwen3.8-Max on August 3, 2026. It is callable through Alibaba's Qwen API with a 1M-token context window, but the launch post did not include a public dollars-per-million-tokens price. Here is what is confirmed, what remains unknown, and what the rest of the Qwen line costs today.

The short answer

  • No confirmed public per-token price yet. The August 3 release documents API access but publishes no input or output rate. We will not infer a price from subscription credits.
  • API access is live. Qwen documents OpenAI-compatible chat-completions and responses APIs, plus an Anthropic-compatible endpoint for agent frameworks.
  • 1M context and 65,536 max output tokens. The official Codex and OpenClaw configuration examples publish these limits for qwen3.8-max.
  • Open weights are due next week. Qwen says this will be its first Max-scale open-weight release; the license is still not stated in the launch post.

What Alibaba confirmed

Released
August 3, 2026
Parameters
2.4 trillion
Activated parameters
95 billion
Context window
1,000,000 tokens
Modalities
Text, images, video and documents — the first Qwen above 1T parameters to go multimodal
Availability
Qwen API, Qwen Studio and compatible agent frameworks
Reasoning controls
xhigh (default), medium and low

Confirmedspecs These values now come from Qwen's August 3 launch post and its published API configuration. The extensive benchmark table is vendor-reported; independent evaluations should still be treated separately.

What the subscription costs

Alibaba launched an individual tier of Token Plan alongside the preview. Prices below are the reported China list prices, in RMB per month, with the USD figures converted at the time of reporting — they are Unconfirmedas reported, not taken from a vendor price sheet we could reach.

Individual tierPer monthApprox. USD
Lite¥39~$5.80
Standard¥139~$20.50
Pro¥499~$73.70

Team seats are reported at ¥150, ¥550 and ¥1,398 per seat per month. On top of the tier price, credit consumption is discounted 90% during daytime hours, and individual subscribers get a further 80% off at night — so the effective rate swings by an order of magnitude depending on when your jobs run. That promotional structure is the reason a stable per-token number does not exist yet.

Qwen models you can price per token

If you are costing out a build today, these are the Qwen models with real published rates. Qwen3.7 Max is the closest thing to a proxy for where 3.8 lands — it is the outgoing flagship, at $1.475 in and $4.425 out per 1M tokens.

ModelInput / 1MOutput / 1MContext
Qwen3.8-MaxNot published — subscription only1M
Qwen3.7 Max$1.475$4.4251M
Qwen3.7 Plus$0.320$1.2801M
Qwen3.5 Plus$0.400$2.4001M
Qwen3.5 Flash$0.065$0.2601M

Rates as listed on OpenRouter, retrieved July 20, 2026. Alibaba Cloud's own Model Studio rates can differ by region and by volume tier.

Cost calculator

Enter monthly volume and the typical prompt size. The applicable context tier is selected automatically.

TierInput costOutput costMonthly total
Qwen3.7 Max$1.48$0.885$2.36
Qwen3.7 Plus$0.32$0.256$0.576
Qwen3.5 Plus (Alibaba international, ≤256K)$0.40$0.48$0.88
Qwen3.5 Flash$0.065$0.052$0.117cheapest

Estimates only — list prices, before any caching discount, batch pricing, or enterprise agreement.

What to watch for

Two things would turn this page into a complete cost reference: an official regional pay-as-you-go price table and next week's promised open-weight release. The latter will make self-hosting possible, but hardware and serving costs cannot be estimated until the weight format and license are published. Until then, the honest answer is that API access is confirmed while per-token cost is not.

Also on this site: Kimi K3 pricing — the model Qwen3.8 was previewed three days after, and the one Chinese flagship that does publish a per-token rate — and GPT-5.6 pricing, the model Qwen3.8 is being benchmarked against on cost.

Sources

Figures on this page last checked against these sources on 2026-08-03. Vendors change pricing and specs without notice — if a number here disagrees with the vendor's own page, trust the vendor.