Qwen3.5 397B A17Bon Krater.ai

The Qwen3.5 series 397B-A17B native vision-language model is built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency.

Released February 16, 2026

Specifications

Qwen3.5 397B A17B at a glance

Context

262.1K tokens

Max output

65.5K tokens

Input

Text, Image, Video

Output

Text

Input price

$0.39 / 1M

Output price

$2.34 / 1M

Image inputTool callingStructured outputsReasoning support

Capabilities

What it is good at

Catalog capabilities

  • Accepts image input
  • Generates text responses
  • Supports tool calling
  • Supports structured outputs

Good for

Image understandingLong documentsUse with toolsStructured responsesDeep analysis

Benchmarks

Catalog data, not Krater benchmarks.

Coding index score

48.2

Agentic index score

19.8

Intelligence index score

34.3

Design Arena svg rank

#35

Design Arena 3d rank

#43

FAQ

Questions about Qwen3.5 397B A17B

Clear answers based on the model catalog and Krater access.

Qwen3.5 397B A17B supports a context window of 262,144 tokens.

Use Qwen3.5 397B A17B in your API

Connect with Krater's unified API.

API access

Ready to use Qwen3.5 397B A17B?

Access leading models in one focused workspace.