Alibaba Claims Qwen4-Max Matches US Frontier Models
Alibaba's Qwen4-Max packs 2.4 trillion parameters and a 1-million-token context window. The claim it rivals OpenAI rests on self-reported benchmarks against Chinese rivals, not US models.
By Daniel Okafor
3 min read

What's News
- Qwen4-Max is a 2.4-trillion-parameter MoE model with a 1-million-token context window
- The API costs $1.50/$4.50 per million input/output tokens and is served from Singapore
- Alibaba claims up to ~50% fewer tokens than Qwen3.8-Max for comparable results
- Benchmarks compare against Kimi K3.1 Pro and Fable 5.1, not US frontier models
- Alibaba has pledged more than $53 billion over three years for AI; T-Head filed for a Hong Kong IPO in September
By Daniel Okafor
Alibaba Group has released Qwen4-Max, a 2.4-trillion-parameter mixture-of-experts model with a 1-million-token context window, and claims performance that rivals Anthropic and OpenAI. The claim rests entirely on benchmarks Alibaba ran itself.
Here is what the launch actually contains, and what it does not.
What did Alibaba release?
Qwen4-Max is a closed model. It is not open-weight. According to a Bluesky post cited by Techmeme, the API is served from Singapore. It costs $1.50 per million input tokens and $4.50 per million output tokens, per industry convention.
The model has one measurable improvement claim attached to it. Alibaba says Qwen4-Max beats its predecessor, Qwen3.8-Max, while using up to roughly 50% fewer tokens for comparable results.
That figure needs a caveat. Alibaba did not specify the task distribution on which the token savings were measured. "Up to" is doing heavy lifting in that sentence.
Alibaba also said it plans to release weights for smaller Qwen4 models. No timeline was given.
How strong are the benchmark claims?
Alibaba claims Qwen4-Max outperforms Fable 5.1 and Kimi K3.1 Pro on coding benchmarks. The company frames the result as evidence that its flagship now rivals Anthropic PBC and OpenAI, in what Techmeme describes as China's latest challenge to US leaders.
Look closely at the comparison set. The benchmarks Alibaba cites compare Qwen4-Max against Kimi K3.1 Pro, a Chinese model from Moonshot AI, and Fable 5.1. They do not compare it against US frontier models from Anthropic or OpenAI.
So the parity claim has a gap in it. Alibaba benchmarks against Chinese rivals, then asserts parity with American labs. Without third-party replication on standard suites, the claim remains unverified. That is the single most important fact for any buyer or analyst reading this launch.
Why does the launch matter geopolitically?
The release arrives amid US export controls that restrict China's access to foundry capacity. Those controls shape how Chinese labs build and deliver models.
Alibaba has responded with money and silicon. The company has invested heavily in domestic chip development through its T-Head unit, which filed for a Hong Kong IPO in September, according to reporting cited from briefs.co. Alibaba has also pledged more than $53 billion over three years to expand its AI capabilities, per Bloomberg.
The Singapore API serving point fits the same pattern. Delivering from a jurisdiction outside mainland China reflects what Techmeme calls the geopolitical contours of the current AI race.
What is still unknown?
The available reporting is silent on several things a serious evaluation would require:
- Training compute
- Data sources
- Energy consumption
- Safety evaluations
- Deployment details
None of these appeared in the launch materials as reported. Buyers pricing the $1.50/$4.50 API against US alternatives are working with vendor-reported numbers and no safety record.
What is the so-what?
If independently verified, Qwen4-Max would represent the closest a Chinese lab has come to matching US frontier capabilities on paper. Until then, it is a claim backed by self-reported benchmarks against selected rivals, priced competitively, and served from Singapore rather than Hangzhou. The next data point that matters is third-party evaluation on standard suites — and Alibaba's delivery date for the open-weight smaller Qwen4 models it has promised but not scheduled.
Source: epyhia.com
More from Daniel Okafor
Show full bio
Correspondent covering money, funding and market shifts around entrepreneurship, startups and venture funding at Business Bearings.
2,189 articles