Skip to navigation
Models

Qwen

Qwen is Mixlayer’s general-purpose model family. It spans a free small model through larger models for difficult reasoning and agentic work.

Model details

  • Context window: 131K tokens, or 262K for Qwen 3.8 27B
  • Capabilities: Text · Vision · Reasoning · Tools

Available models

ModelIdentifier
Qwen 3.5 4B (Free)
Free prototyping and short, low-stakes tasks.		qwen/qwen3.5-4b-free
Qwen 3.5 9B
High-volume chat, classification, and summarization.		qwen/qwen3.5-9b
Qwen 3.6 35B (3B active)
Stronger reasoning, coding, and agents with MoE efficiency.		qwen/qwen3.6-35b-a3b
Qwen 3.8 27B
General-purpose reasoning, coding, and agents with a 262K context window.		qwen/qwen3.8-27b

Thinking modes

Qwen supports a thinking mode, with reasoning returned in reasoning_content, and a non-thinking mode for faster direct answers. See Reasoning for request examples.

Thinking is enabled by default. Use none to disable it. All other efforts (default, minimal, low, medium, high, xhigh, and max) select the same default thinking behavior. Set the value through reasoning_effort in Chat Completions or reasoning.effort in Responses.

Mixlayer platform defaults

When these parameters are omitted from a request, Mixlayer applies the following defaults:

IdentifierDefault settings
qwen/qwen3.5-4b-freetemperature=1.0, top_p=0.95, top_k=20, presence_penalty=1.5
qwen/qwen3.5-9btemperature=1.0, top_p=0.95, top_k=20, presence_penalty=1.5
qwen/qwen3.6-35b-a3btemperature=1.0, top_p=0.95, top_k=20
qwen/qwen3.8-27btemperature=1.0, top_p=0.95, top_k=20
ModeRecommended sampling
Thinking — general taskstemperature=1.0, top_p=0.95, top_k=20, presence_penalty=0.0, repetition_penalty=1.0
Thinking — precise codingtemperature=0.6, top_p=0.95, top_k=20, presence_penalty=0.0, repetition_penalty=1.0
Non-thinkingtemperature=0.7, top_p=0.80, top_k=20, presence_penalty=1.5, repetition_penalty=1.0

Qwen also recommends min_p=0.0. Mixlayer does not currently expose min_p, so omit it and use the gateway default.

For the mixture-of-experts variant, active parameters primarily govern inference cost and latency while total parameters contribute to capability.