Qwen3.8-27B API.
Code. Reason. Build.

Qwen3.8-27B is Qwen’s 27-billion-parameter dense open-weight model for coding and reasoning. Blade serves it through an OpenAI-compatible API with 64k context, so you can build without managing a GPU deployment.

€5 in free credits for every new customer.

Build with Qwen

Input from €0.0816 · output from €1.09 per million tokens. See pricing details.

Code generation, chat
and reasoning.

Start with a specific job. Supply the right context, define what success looks like and evaluate the result in your own workflow.

Generate code and tests.

Draft a function, suggest a fix or generate regression tests. Send the relevant files and acceptance criteria; run the patch through your tests before merging.

Bug fixes / tests / refactoring

Build a technical assistant.

Build a technical support assistant that works from your documentation, logs or product context. Your backend supplies the evidence and keeps the conversation under your control.

Developer support / product assistants

Plan complex changes.

Compare implementation options or turn a complex issue into a plan. For agent workflows, connect only tools supported by the endpoint and validate actions in your application.

Planning / analysis / coding agents

Coding quality

Qwen3.8-27B
coding benchmarks.

Qwen reports improvements in coding and multi-step tasks. Two results from its official model card make Qwen3.8-27B worth testing on your own codebase.

SWE-bench Pro

Agentic software engineering

61.7

Qwen-reported score using the Claude Code harness, a 256K context window and a corrected task set.

LiveCodeBench v6

Competitive coding

90.3

Qwen-reported score for coding problems. A different test from maintaining an existing application.

Benchmark source: Qwen’s official model card, checked 12 October 2026. These are publisher results, not benchmarks of Blade’s W4A16 (INT4) endpoint with 64k context. Quantization, settings and your task can change the outcome.

Test the patch. Use representative bugs and the same test suite you use in production.

Count the whole job. Compare retries, review time and total tokens, not just the first answer.

Keep the scope clear. A model can suggest code; your agent or application runs tools and checks permissions.

Qwen3.8-27B pricing

Qwen3.8-27B
API pricing.

Pay for the tokens your requests use. Blade’s dynamic pricing moves within a model-specific range: the amounts shown here are catalogue starting rates, not a fixed quote for every request.

See the full rate card
Input / 1M tokens
From€0.0816
Output / 1M tokens
From€1.09

A concrete example
10,000 input tokens + 2,000 output tokens = €0.002996 at the displayed starting rates.

Illustration only; excludes additional requests and paid tools. Catalogue updated 2026-10-22. Confirm current prices and model access in the console.

Check current rates

OpenAI-compatible API

Call Qwen3.8-27B
with the OpenAI SDK.

Create a Blade API key, check that qwen3.8-27b is available to your account, then call it from your backend.

  1. Install the Python client: pip install openai.
  2. Set BLADE_API_KEY in your server environment.
  3. Run the example. This sends a billed inference request.

Keep your API key on the server. For a production integration, handle timeouts, rate limits and truncated responses.

Follow the complete setup guide
python
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.models.blade.sh/v1",
    api_key=os.environ["BLADE_API_KEY"],
    timeout=120.0,
)

response = client.chat.completions.create(
    model="qwen3.8-27b",
    messages=[{
        "role": "user",
        "content": (
            "Write a Python retry function with "
            "exponential backoff. Add pytest tests "
            "that do not actually sleep."
        ),
    }],
    max_tokens=4096,
)

print(response.choices[0].message.content)

Use the Blade model ID, not the upstream repository name. Supported parameters are documented in the API reference.

In your coding workflow

Keep Claude Code or Codex.
Delegate to Qwen.

Use Qwen3.8-27B alongside your coding assistant through a skill or callable script that sends tasks to the Blade API. Your main assistant scopes the work, reviews the proposal and runs your checks.

01 / ScopeClaude Code or Codex

Choose a task and the relevant files.

02 / GenerateQwen3.8-27B

Return a proposed patch via Blade.

03 / ReviewYour coding assistant

Inspect the changes. Run the tests.

Start with one focused task.

Generate tests, draft documentation or refactor an isolated function. Configure the API runner with your Blade key and qwen3.8-27b, then connect it to your assistant using a skill.

Set up coding delegation 
Delegate the retry function's tests to
Qwen through Blade. Share only the function
and its existing tests. Review the proposed
diff and run pytest before applying it.

Requires a configured skill or API tool; pasting this prompt alone does not install the integration. Blade API usage is billed separately. Your primary assistant keeps its existing model.

Integration building blocks: Claude Code skills and Codex skills.

Qwen3.8-27B API FAQ.

What is Qwen3.8-27B?

It is a 27-billion-parameter dense open-weight model from Qwen. On Blade, the catalogue lists it as a text reasoning model with 64k context and W4A16 (INT4) quantization. Use it through the Models API without serving the weights yourself.

Is Qwen3.8-27B good for coding?

Qwen’s published evaluations make it a candidate for code generation and agent-assisted development. They do not establish the success rate on your project. Evaluate bug fixes, tests and refactoring with your own files, review process and constraints.

Does Blade offer the full upstream context and vision features?

The hosted configuration matters. Blade currently lists 64k context for this model. Do not assume every capability of the upstream model is enabled on this endpoint; check supported inputs and parameters in the console before integrating image, video or reasoning controls.

Can I use Qwen3.8-27B with Claude Code or Codex?

Yes, through a delegation skill or API tool configured to call Blade. Claude Code or Codex sends a focused task to qwen3.8-27b, then reviews the returned code. This workflow keeps your assistant’s main model in place and bills the delegated requests through your Blade account. Follow the coding delegation setup.

Can I use it in a coding agent or with web search?

A tool that accepts a custom OpenAI-compatible provider can use Blade’s base URL, API key and model ID for supported calls. Agent behaviour depends on its integration and the endpoint’s capabilities. Flash by Staan search requires account access and a compatible tool-calling model, and is billed separately. See the tool-calling guide before wiring it in.

How much does the Qwen3.8-27B API cost on Blade?

Input starts at €0.0816 per million tokens and output starts at €1.09 per million tokens, according to Blade’s catalogue dated 2026-10-22. These are dynamic pricing floors, not fixed rates. Check the applicable prices in the console and read how inference pricing works.

Compare coding and reasoning options. Browse all models ↗
Need your own runtime? Plan a Qwen3.8-27B GPU deployment ↗

Qwen3.6-35B-A3BCoding & reasoning ↗Laguna XS 2.1Coding & reasoning ↗gpt-oss-20bReasoning & coding ↗

Contact the team

Loading the form…