Summary:

Alibaba’s hosted flagship starts at $1.50 per million standard input tokens in Singapore. Vals test costs and scores show how it compares with Neon and Opus.

Alibaba released Qwen4-Max on November 16, making its new flagship available as qwen4-max through the hosted API. Pricing is one reason to consider it: the Singapore deployment lists standard input at $1.50 and output at $4.50 per million tokens. Independent Vals results point to a more specific reason to try it: Qwen4-Max beat Gemini 4 Neon and Claude Opus 5.5 on one legal-agent test, but trailed both on terminal tasks and code migration.

What is available now?

Qwen4-Max is accessible through QwenCloud’s API and Token Plan, and Alibaba also lists QwenWork and QwenStudio as ways to try it. Model Studio documents compatibility with OpenAI Chat Completions and Responses, Anthropic Messages and DashScope. The model takes text, images and video as input and returns text. Alibaba describes it as a million-token-context model with 2.4 trillion total parameters and 71 billion active per token.

The hosted Max release is distinct from the smaller Qwen4 models Alibaba says it plans to open-weight. The availability of another model in the family does not make Max downloadable.

Qwen4-Max API pricing

For Model Studio’s Singapore deployment, Alibaba lists US$1.50 per million standard input tokens and US$4.50 per million output tokens. The output price covers reasoning and the response. At those standard rates, 100,000 input tokens and 10,000 billed output tokens cost about $0.20 before other charges or discounts. Singapore is listed as international; Model Studio documents other deployments separately. Check the pricing and feature table for the endpoint you intend to use.

Qwen4-Max versus Gemini 4 Neon and Claude Opus 5.5

These rows come from the independent Vals Index v2.2 leaderboard dated November 18, 2026. All values are accuracy scores, and the highest within these three models is bolded. The Code Migration row uses the subset included in the index, not the full standalone test.

BenchmarkQwen4-MaxGemini 4 NeonClaude Opus 5.5
Vals Index v2.2 overall63.20%70.87%70.14%
Finance Agent v2.166.7%68.6%67.8%
Terminal-Bench 4.048.5%61.6%65.2%
Code Migration v1.1 (index subset)57.3%75.8%80.5%
HLAB-Vals-1.121.3%19.6%9.2%

Qwen4-Max scored 63.20% overall, up from 59.05% for Qwen 3.8 Max 0902 on the same leaderboard. In the three-model comparison, it scores highest on Harvey’s legal-agent benchmark; the other two lead on the selected finance and coding tests. Vals also reports $9.33 per index test for Qwen4-Max, versus $7.40 for Gemini 4 Neon and $24.77 for Claude Opus 5.5. Those are costs on Vals’ test mix, not API quotes. A lower token rate can still yield a higher cost per job if a model uses more tokens.

If you are choosing between these models, measure cost per successful task and time to completion on the same inputs. For terminal-heavy work, the two coding rows are more informative than the aggregate index. For a closer look at Google’s option, see our Gemini 4 Neon pricing and benchmarks breakdown.

Sources