Skip to main content
FlagshipAlibaba

Qwen 3.8 Max

Qwen 3.8 Max is Alibaba's flagship-tier model in the Qwen 3.8 family, accepting text, image, and video input with a 1M-token context window and tool calling support.

Context 1.0M
Tier Flagship
Tools Supported
Modalities text, image, video
Input from
$1.65 / 1M tokens
across 7 providers

API Pricing

Cheapest on Deep Infra — 13% below avg
ProviderInput / 1MOutput / 1MCached / 1M
$1.65$4.95$0.206
$1.65$4.95$0.206
$1.90$5.70-
$2.00$6.00$0.200
$2.00$6.00-
$2.00$6.00$0.250
$2.00$6.00$0.250

Prices updated daily. Last check: Sep 24, 2026

Qwen 3.8 Max pricing by provider

Input, output, and batch rates, plus alternatives, for one provider at a time.

Performance & Benchmarks

Source: Artificial Analysis →
Intelligence
45.4 / 100
Coding
76.2 / 100
Output Speed
39.0 t/s
Latency (TTFT)
2.0s

Reasoning & Knowledge

  • GPQA Diamond92.8%
  • Humanity's Last Exam43.1%

Coding

  • SciCode52.1%

Agentic & Tool Use

  • Terminal-Bench v2.188.8%
  • τ-bench Banking47.8%

Instruction & Long Context

  • Long-Context Reasoning80.3%

Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.

Model Details

General

Creator
Alibaba
Family
Qwen 3.8
Tier
Flagship
Context Window
1.0M
Modalities
Text, Image, Video

Capabilities

Tool Calling
Yes
Open Source
No
Aliases
qwen3-8-max, qwen/qwen3.8-max, qwen3.8-max, Qwen3.8-Max

Strengths & Limitations

Strengths

  • 1,000,000-token context window, enough for entire codebases, long document sets, or large transcript archives in a single prompt
  • Accepts image input in addition to text, allowing visual documents, screenshots, and charts to be reasoned over directly
  • Accepts video input, which is less common than image support among API-served models
  • Tool calling support enables use in function-calling and agentic pipelines
  • Flagship tier of the Qwen 3.8 family, so it is the capability target within that generation rather than a distilled or cost-reduced variant
  • Exposed under multiple provider aliases (qwen3-8-max, qwen/qwen3.8-max, Qwen3.8-Max), making it straightforward to switch hosts
  • Independently benchmarked throughput and latency figures available from Artificial Analysis for capacity planning

Limitations

  • Measured output speed of roughly 47 tokens per second is modest, which is noticeable in streaming chat or interactive coding sessions
  • Time to first token around 1.8 seconds adds perceptible lag for latency-sensitive interfaces
  • Flagship-tier pricing generally makes it a poor fit for high-volume, low-complexity traffic that a smaller Qwen model could handle
  • Filling anywhere near the 1M-token window is expensive and increases latency, so the full context is rarely practical for routine requests
  • We do not track detailed public benchmark scores for reasoning, coding, or math for this model, so capability comparisons against peers must rely on your own evaluations

Key Features

•1,000,000-token context window
•Text input and output
•Image (vision) input
•Video input
•Tool / function calling
•Flagship tier of the Qwen 3.8 family
•Multiple provider aliases for portable integration
•Third-party performance measurements from Artificial Analysis

About Qwen 3.8 Max

Qwen 3.8 Max is a large language model from Alibaba and sits at the flagship tier of the Qwen 3.8 family. It is the Max-branded variant of the family, positioned above smaller Qwen releases intended for lighter or cheaper workloads. It is served through API endpoints and appears under several identifiers across providers, including qwen3-8-max, qwen3.8-max, and qwen/qwen3.8-max. The model has a 1,000,000-token context window and accepts text, image, and video input, making it usable for long-document work and for prompts that mix written material with visual media. Tool calling is supported, which allows the model to be wired into function-calling and agent frameworks that require structured invocation of external APIs. Independent measurements from Artificial Analysis put throughput at roughly 47 output tokens per second with time to first token around 1,806 ms; both figures depend heavily on the serving provider and load, so treat them as a reference point rather than a guarantee. In practice, a flagship-tier model with a million-token window and multimodal input tends to be used where the input itself is the hard part: entire codebases, long contract or filing sets, transcript archives, or video and image material that needs to be summarized or cross-referenced. Compared with lower-tier Qwen models, the trade-off is the usual one — more capability and a larger context ceiling in exchange for higher per-token cost and slower generation, so many teams route only the difficult subset of traffic to it.

Common Use Cases

Qwen 3.8 Max is aimed at workloads where context size or input modality is the binding constraint. Typical fits include whole-repository code analysis and refactoring plans, review of long legal or financial document bundles, multi-document research synthesis, and question answering over large transcript or log collections. The image and video inputs extend this to visual workloads — describing or indexing video content, extracting data from scanned or screenshotted material, and answering questions that mix visual and textual evidence. Tool calling makes it viable as the planning model in an agent stack that queries databases or external APIs. Given its flagship-tier cost profile and mid-range generation speed, it is most economical in an architecture where a cheaper Qwen model handles routine classification, extraction, and short-form chat, and Qwen 3.8 Max is reserved for the long-context or multimodal requests that smaller models cannot serve.

Frequently Asked Questions

How much does Qwen 3.8 Max cost?

Pricing varies by provider and by pricing model — input and output tokens are usually billed at different rates, and some hosts apply separate rates for image or video input or for very long contexts. Check the pricing table on this page for current per-provider rates.

What is Qwen 3.8 Max best used for?

It suits long-context and multimodal work: analyzing entire codebases, reviewing large document bundles, synthesizing across many sources, and answering questions over image or video input. Its tool calling support also makes it usable as the reasoning step in agent workflows.

How large is the context window?

Qwen 3.8 Max supports a 1,000,000-token context window. Note that some providers cap the usable context below the model maximum, and filling the window raises both cost and latency, so check your provider's documented limit.

Does Qwen 3.8 Max support images and video?

Yes. Our metadata confirms text, image, and video input modalities. Video support in particular is less common among API-served models, though the specific formats, resolutions, and duration limits accepted depend on the provider.

How fast is Qwen 3.8 Max?

Artificial Analysis measurements put output throughput at about 47 tokens per second with time to first token near 1,806 ms. Both numbers vary by provider, region, prompt length, and current load, so treat them as a baseline rather than a guarantee.

Should I use Qwen 3.8 Max or a smaller Qwen model?

Use Qwen 3.8 Max when you need the flagship-tier capability, the million-token context, or video and image input. For high-volume classification, extraction, or short chat turns, a smaller model in the Qwen line is generally cheaper and faster; many deployments route between the two.