Qwen3.8 Max: Benchmarks, Pricing, Context & API

Alibaba · Proprietary

Alibaba’s hosted quality ceiling: strong on agentic professional work, but slower in current observations than the open-weight A95B sibling. Source-checked August 28, 2026.

Compare this model Open live ranking

Live rank: #7
Intelligence v4.1: 58.1
Best blended / 1M: $3.00
Context: 1M

The answer in 60 seconds

Choose Max for the managed ceiling, not for openness or speed

Qwen3.8 Max is the hosted Qwen quality tier. Its current independent scores are strongest on agentic work, and Alibaba’s managed API removes the infrastructure burden of the enormous open-weight sibling. The measured trade-off is throughput: current WhatLLM observations are much slower than A95B and the fastest Flash-class alternatives, so interactive products should test latency with real output lengths.

Best for

— Managed Qwen deployments in Alibaba Cloud
— Agentic knowledge work and coding where hosting simplicity matters
— Teams that want the strongest Qwen3.8 endpoint without running multi-trillion-parameter weights

Look elsewhere when

— Open weights or deployment portability is a hard requirement
— High output throughput is central to the user experience

Independent data

What the current measurements say

Metrics come from the live Artificial Analysis-backed WhatLLM feed and refresh hourly. Prices and speed are best observed values across matching endpoints; they can come from different providers.

Intelligence Index: 58.1
Coding Index: 71.8
Agentic Index: 58.4
Input / output per 1M: $2.00 / $6.00
Fastest observed: 25.7 tok/s
Lowest observed TTFT: 1.756s

Benchmarks worth reading together

The strongest indexed configuration is shown. Different tests measure different jobs; no single score is a universal model verdict.

GDPval-AA Elo: 1724
GDPval-AA: 61%
GPQA Diamond: 93%
Humanity's Last Exam: 43%
SciCode: 53%
AA-LCR: 74%

Source truth

What Qwen3.8 Max actually is

Access: Alibaba Cloud Model Studio hosted API; Provider-confirmedSource ↓
Context: 1 million tokens; Provider-confirmedSource ↓
Modalities: Text, image, and video input; text output; Provider-confirmedSource ↓
Thinking and tools: Non-thinking support plus managed tools; Provider-confirmedSource ↓
Weights: Not released for the Max endpoint; Provider-confirmedSource ↓

Editorial analysis

The decision behind the ranking

We have not claimed hands-on use here. The analysis below combines provider-confirmed facts with the independent data shown above; each section labels that distinction.

WhatLLM analysis

Max and A95B answer different ownership questions

The Max API buys a managed service and Alibaba’s highest Qwen3.8 score. The 2.4T A95B release buys weights, deployment control, and a near-frontier score at the cost of a formidable serving problem. The small score gap is not the decision; governance, operations, and throughput are.

Teams already on Model Studio can start with Max. Teams evaluating sovereign or inspectable deployment should benchmark A95B through a specialist host before committing to self-hosting.

Independent signal

Throughput is the live caveat

The current fastest observed Max endpoint trails several peers by a wide margin in output tokens per second. That may change as providers tune serving, but it is material today. Test time-to-first-token and full response time separately; a good first token does not guarantee a fast long answer.

Boundaries

What the headline score misses

01The Max weights are closed.
02Current observed output throughput is comparatively low.
03The strongest official claims come from Alibaba’s own launch evaluation, and no dedicated Qwen3.8 safety card surfaced.
04A one-million-token capacity is not evidence of uniform retrieval quality at that length.
05Availability and pricing vary by Alibaba Cloud region.

Alternatives

The next models to test

Compare the shortlist →

Qwen3.8 2.4T A95B

Near-Max Qwen capability with downloadable weights—and a 2.4T-scale infrastructure bill hidden behind the word “open.”

GLM-5.3

A low-priced, agentic coding specialist whose strongest release is hosted rather than open weight.

GPT-5.6 Terra

The balanced GPT-5.6 tier: much of Sol’s coding capability at half the published token price and materially higher throughput.

Evidence register

Sources and claim status

Checked August 28, 2026. Pricing and availability should be verified again before procurement.

Direct answers

Questions people ask about Qwen3.8 Max

Is Qwen3.8 Max open weight?

No. Qwen3.8 Max is a managed proprietary endpoint. Qwen3.8 2.4T A95B is the open-weight release in the same generation.

What is the Qwen3.8 Max context window?

Alibaba documents a one-million-token context window for the hosted model.

Should I use Qwen3.8 Max or 2.4T A95B?

Use Max when managed access and the highest Qwen score matter. Use A95B when weight access, deployment control, or provider choice justifies the operational complexity.

Assembling the frontier