Qwen4-Max Is Live on QwenCloud at $1.50/$4.50 per 1M Tokens, 25% Below Qwen3.8-Max
- What happened
- Qwen4-Max is live on QwenCloud at $1.50/$4.50 per 1M tokens with a 1M context.
- Why it matters
- That is 25% below Qwen3.8-Max's $2/$6 on the same API, and cached input drops from $0.25 to $0.075.
- What to do
- Run your own evals on qwen4-max before moving qwen3.8-max traffic; we could not verify any benchmark.
Our verdict: if you call qwen3.8-max on QwenCloud today, put qwen4-max on your test list this week. It is cheaper on every list line we read, it sits on the same QwenCloud API, and QwenCloud says it is compatible with both the OpenAI and Anthropic APIs. Do not migrate on launch claims: we could not verify any benchmark. Our Qwen3.8-Max rating stays conditional, and Qwen4-Max is not yet in our directory.
What happened
QwenCloud's model page lists Qwen4-Max (qwen4-max), marked NEW, with a 1M-token context, text, image and video input and text output, aimed at coding, productivity and agentic tasks. QwenCloud says it works with both the OpenAI and Anthropic APIs. Alibaba's launch post on the Qwen blog did not load for us, so we do not repeat its parameter counts, availability list or open-source plans.
Both price rows below come from QwenCloud's own model pages, read on 17 November.
| Per 1M tokens | Qwen4-Max (qwen4-max) | Qwen3.8-Max (qwen3.8-max) |
|---|---|---|
| Input | $1.50 | $2.00 |
| Output | $4.50 | $6.00 |
| Implicit-cache input | $0.075 | $0.25 |
| Explicit cache read | $0.075 | $0.17 |
| Context window | 1M | 1M |
| Max output | 393K | 131K |
| Tokens per minute | 5,000,000 | 2,000,000 |
Why it matters
List input and output are both 25% lower (our arithmetic). The bigger cut is on cached input, which is where long agent sessions spend: implicit-cache input drops 70%, from $0.25 to $0.075, and explicit cache reads fall from $0.17 to $0.075. The rate limit also rises from 2M to 5M tokens per minute.
We could not verify any benchmark: Alibaba's launch post did not load for us, so we report no scores. Until independent numbers exist, the list price cut, which comes straight from QwenCloud's rate card, is the only verified reason to test.
What changes for you
- Existing Qwen3.8-Max users: re-run your eval set against
qwen4-maxbefore moving traffic. Same API, so the switch is a model string. - Agent workloads with long contexts: recalculate cached-input cost at $0.075 per 1M. This is where the saving is largest.
- Batch users: QwenCloud's Qwen4-Max page says batch processing is supported but lists no batch price, so do not budget for a discount.
- Self-hosters: QwenCloud lists Qwen4-Max as a hosted API model, and we read no open-weight release. Do not plan around weights.
The model is still listed and billable on QwenCloud, and Alibaba has posted no retirement date. A newer, cheaper sibling does not make the old model worse at what it already does, but it does remove the reason to start new work on it. We have not tested Qwen4-Max ourselves.
What to do
- 1 Re-run your qwen3.8-max eval set against qwen4-max on QwenCloud before switching traffic.
- 2 Recalculate cached-input costs for long agent sessions at $0.075 per 1M.
- 3 Do not assume a batch price or open weights for Qwen4-Max; neither is listed.
Affected tools & models
Never need to catch up again
The weekly delta: only verdict changes and act-now items. No digest filler.