Qwen3.8 27B free API: where to use it for free

Where Qwen3.8 27B is free

Qwen publishes the weights under Apache 2.0; hosted access is free on several providers' free tiers, each with its own id and limits.

Listed, but not free

• OpenRouter (qwen/qwen3.8-27b): paid only; the :free variant was withdrawn.
• Cerebras (qwen-3.8-27b): trial credits only; Cerebras has no permanent free tier.

What Qwen3.8 27B is

Qwen3.8 27B is a dense 27-billion-parameter vision-language model from the Qwen3.8 open-weight family, built for coding, research and long multi-step agent tasks, with thinking on by default and switchable per request.

Parameters

27B, dense

Context window

262,144 tokens native, up to 1M with YaRN

Input

Text, image, video

Output

Text

Reasoning

Thinking on by default, can be turned off per request

Tool calling

Built into the official chat template

Before you rely on it

• Hosted context is below the 262,144-token native window: 131,072 tokens on Groq.
• Groq lists it as a preview model, which can change or be removed at short notice.
• Free tiers are for building and testing; check each provider's data terms before sending sensitive prompts.

Use Qwen3.8 27B with automatic failover

A single free tier runs out. Ficelle puts the free models of every provider you configure behind one local OpenAI-compatible endpoint. Call a routing profile and Qwen3.8 27B is one of the models it can pick; when it rate-limits or fails, the next eligible free model answers. To pin Qwen3.8 27B alone, call the id that ficelle models lists for it, knowing a pinned model has no failover.

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8646/v1", api_key="ficelle-local")

# A routing profile keeps failover: Qwen3.8 27B is one of the free
# models it can pick, and the next eligible one answers if it fails.
client.chat.completions.create(
    model="ficelle/auto-coding",
    messages=[{"role": "user", "content": "..."}],
)

Routing profiles that fit what Qwen3.8 27B is built for:

• Coding assistants ficelle/auto-coding
• Tool calling ficelle/auto-tools
• Vision ficelle/auto-vision
• Reasoning ficelle/auto-reasoning

Whether a profile routes to Qwen3.8 27B on your install depends on your providers' live catalogs and Ficelle's eligibility checks, decided at request time.

Questions