All models
Model card

Qwen 3.8 27B

lx1-qwen3.8-27bGeneral purposeLive in the catalog

The small open-weight sibling of Qwen 3.8 Max: a 27-billion-parameter dense vision-language model that reads a screenshot as readily as a prompt, thinks before it answers, and still comes back fast enough for interactive work. A 262K window covers a long agent session or a sizeable repo. Input is priced like a workhorse; output is dearer, so spend it on turns that look at something or need a careful answer rather than on glue.

Context window262K tokens
Input · per Mtok$0.45
Output · per Mtok$3.2
Served byLayer X1 engine
Where it earns its keep[01/03]
  • 01Native vision on a compact dense model
  • 02262K context for long sessions and repos
  • 03Tools and reasoning without a flagship input bill
Capabilities
Tool callingYes

Strict, schema-faithful tool calls, enforced by the engine on every request — safe to build an agent loop on.

ReasoningYes

Thinks before it answers. Budget max_tokens generously — hidden reasoning counts against it.

VisionYes

Reads images — screenshots, diagrams, and UI states — inline in the conversation.

Behind the endpoint[02/03]

One endpoint. Served by our engine.

Qwen 3.8 27B is served through the Layer X1 engine — zero-downtime serving is the design target, not a status-page apology. You request it by name; everything else is our problem.

Call it by name
curl https://api.layerx1.com/v1/messages \
  -H "x-api-key: lx1_your_key" \
  -H "content-type: application/json" \
  -d '{
    "model": "lx1-qwen3.8-27b",
    "max_tokens": 1024,
    "messages": [{ "role": "user", "content": "Hello" }]
  }'

OpenAI-style clients work too — send the same model name to /v1/chat/completions with a Bearer key. See the docs for both dialects.

Put your agent on inference built for the work

Your agent stays the same.
Its inference gets better.

$export ANTHROPIC_BASE_URL=https://api.layerx1.com

Start free · no card · Starter from $5/mo