NEAR AI
Sign inStart building
All models

Qwen 3.8 27B Abliterated API

Private TEE
Verified
Qwen/Qwen3.8-27B-abliterated

Qwen 3.8 27B with refusal behavior removed (abliterated), for red teaming, security research and evaluations. Reasoning, tool use, vision and a 262K context window. Served privately from NEAR AI's attested TEE infrastructure.

Input / 1M tokens
$0.04
Output / 1M tokens
$1.35
Cache read / 1M tokens
$0.018
Context window
262K tokens

Privacy & execution

This model supports private inference inside a hardware-isolated Trusted Execution Environment (TEE), protecting data while it is processed. HTTPS protects data in transit; client-to-model end-to-end encryption depends on the model and integration.

Attestation provides cryptographic evidence of the execution environment’s identity and configuration. Where response signatures are available, you can also verify the integrity of the request and response.

Verification documentation
01

Make your first request

Call /v1/chat/completions over HTTPS.

Get an API key and set the NEARAI_API_KEY environment variable.

curl --fail-with-body "https://cloud-api.near.ai/v1/chat/completions" \ -H "Authorization: Bearer $NEARAI_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "Qwen/Qwen3.8-27B-abliterated", "messages": [ { "role": "user", "content": "Hello!" } ] }'

This HTTP example uses HTTPS. Use the SDK below to add field-level end-to-end encryption and attestation checks.

02

Build and Verify with the SDK

Let the inference client handle verified transport and message encryption.

Verification guide

Set NEARAI_API_KEY before running an example.

Setup

npm install @nearai/inference-sdk openai

Example

import OpenAI from "openai"; import { InferenceClient } from "@nearai/inference-sdk/node"; const apiKey = process.env.NEARAI_API_KEY; if (!apiKey) throw new Error("Set NEARAI_API_KEY first."); // Verify Gateway and model attestations, and encrypt supported fields to the model. const inferenceClient = new InferenceClient({ apiKey, baseUrl: "https://cloud-api.near.ai/v1", e2ee: true, }); const openai = new OpenAI({ apiKey, baseURL: inferenceClient.getBaseUrl(), fetch: inferenceClient.fetch, maxRetries: 0, }); const completion = await openai.chat.completions.create({ model: "Qwen/Qwen3.8-27B-abliterated", messages: [{ role: "user", content: "Hello!" }], }); // Verify the exact request/response bytes before displaying the answer. const verified = await inferenceClient.verifyResponse(completion.id); console.log(completion.choices[0]?.message.content ?? ""); console.log("Verified signer:", verified.signatureKind);

JavaScript SDK guide

What does abliterated mean?

Qwen 3.8 27B Abliterated is an abliterated model: the internal direction that makes a chat model refuse has been removed from its weights, without retraining. It answers the security questions mainstream models refuse, such as how your own code could be attacked.

Qwen 3.8 27B Abliterated API FAQ

How much does the Qwen 3.8 27B Abliterated API cost?
$0.04 per 1M input tokens and $1.35 per 1M output tokens on NEAR AI Cloud, and $0.018 per 1M cached input tokens.
What is the context window of Qwen 3.8 27B Abliterated?
262K tokens.
How do I call the Qwen 3.8 27B Abliterated API?
Create an API key on NEAR AI Cloud, then send requests to https://cloud-api.near.ai/v1/chat/completions with the model ID Qwen/Qwen3.8-27B-abliterated. The endpoint is OpenAI-compatible, so OpenAI SDKs work by setting the base URL to https://cloud-api.near.ai/v1.
What is the Qwen 3.8 27B Abliterated API used for?
Security work: finding vulnerabilities in your own code, authorized testing, triage and red-teaming your own AI systems. An abliterated model answers security questions that a standard model may refuse. Get an API key on NEAR AI Cloud and call it like any OpenAI-compatible model.
Is Qwen 3.8 27B Abliterated private on NEAR AI?
Yes. It runs inside a hardware-isolated Trusted Execution Environment (TEE) on NEAR AI's infrastructure, which protects data while it is processed. Attestation provides cryptographic evidence of the execution environment, which you can verify.