Zero‑Cost Unlimited‑Token Access to Qwen 3.6: A Step‑by‑Step Guide

Model overview

Qwen3.6-35B-A3B is an open‑source Mixture‑of‑Experts (MoE) model released in April 2026. It has 350 billion total parameters but activates only 30 billion per inference by selecting 8 of 256 expert networks plus one shared expert for each token.

Terminal‑Bench 2.0: 51.5 (vs Qwen3.5‑27B 41.6, Gemma4‑31B 42.9)

SWE‑bench Verified: 73.4 (vs Qwen3.5‑35B‑A3B 70.0)

NL2Repo code‑generation: 29.4 (vs Qwen3.5‑35B‑A3B 20.5)

Despite the reduced active parameters, the model matches or exceeds dense models on programming and agent benchmarks, supports multimodal input and a fast “non‑thinking” mode.

Free unlimited token access via iFlytek MaaS

iFlytek’s MaaS platform provides unlimited free API calls (0 CNY per million tokens) for Qwen3.6‑35B‑A3B and Qwen3.5‑35B‑A3B until 30 June 2026. No daily caps or hidden limits.

Acquisition steps

1. Open the model square

Navigate to the model square URL:

https://maas.xfyun.cn/modelSquare?ch=MaaS-jgkol-f3D8i

2. Locate the free models

Find the cards for Qwen3.6-35B-A3B and Qwen3.5-35B-A3B and click “API调用”.

3. Create an application

Click “前往创建应用”, enter a name, and obtain an appID. The dialog shows three required values:

API Base URL (OpenAI‑compatible endpoint)

API Key (authentication token)

modelId (e.g., xopqwen36v35b)

4. Authorise the app

Select the newly created app, complete real‑name verification, and the service appears in the model list.

Java client example

Add Maven dependencies (or Gradle equivalents) for OkHttp 4.12.0 and Jackson 2.16.1:

<dependency>
  <groupId>com.squareup.okhttp3</groupId>
  <artifactId>okhttp</artifactId>
  <version>4.12.0</version>
</dependency>
<dependency>
  <groupId>com.fasterxml.jackson.core</groupId>
  <artifactId>jackson-databind</artifactId>
  <version>2.16.1</version>
</dependency>

Full source (angle brackets escaped):

import com.fasterxml.jackson.databind.ObjectMapper;
import com.fasterxml.jackson.databind.node.ArrayNode;
import com.fasterxml.jackson.databind.node.ObjectNode;
import okhttp3.*;
import java.io.IOException;
import java.util.concurrent.TimeUnit;

public class QwenMaaSDemo {
    // Configuration – replace with your values
    private static final String API_BASE_URL = "https://maas-api.cn-huabei-1.xf-yun.com/v2";
    private static final String API_KEY = "YOUR_API_KEY";
    private static final String MODEL_ID = "YOUR_MODEL_ID"; // e.g., xopqwen36v35b

    private static final OkHttpClient client = new OkHttpClient.Builder()
            .connectTimeout(60, TimeUnit.SECONDS)
            .readTimeout(60, TimeUnit.SECONDS)
            .build();
    private static final ObjectMapper mapper = new ObjectMapper();

    public static void main(String[] args) throws IOException {
        callQwen("用一句话介绍一下 MoE（混合专家）模型架构", "你是一个专业的AI助手");
        callQwen("写一个 Java 函数，实现归并排序算法，并添加详细的注释", "你是一名资深 Java 工程师，代码风格简洁规范");
    }

    /** Call iFlytek MaaS Qwen model API */
    public static String callQwen(String prompt, String systemPrompt) {
        // 1. Build OpenAI‑style JSON payload
        ObjectNode root = mapper.createObjectNode();
        root.put("model", MODEL_ID);
        root.put("temperature", 0.7);
        root.put("max_tokens", 2048);
        ArrayNode messages = mapper.createArrayNode();
        ObjectNode systemMsg = mapper.createObjectNode();
        systemMsg.put("role", "system");
        systemMsg.put("content", systemPrompt);
        messages.add(systemMsg);
        ObjectNode userMsg = mapper.createObjectNode();
        userMsg.put("role", "user");
        userMsg.put("content", prompt);
        messages.add(userMsg);
        root.set("messages", messages);
        String jsonBody;
        try {
            jsonBody = mapper.writeValueAsString(root);
        } catch (Exception e) {
            System.err.println("❌ JSON construction failed: " + e.getMessage());
            return null;
        }
        // 2. Build HTTP request
        Request request = new Request.Builder()
                .url(API_BASE_URL + "/chat/completions")
                .addHeader("Content-Type", "application/json")
                .addHeader("Authorization", "Bearer " + API_KEY)
                .post(RequestBody.create(jsonBody, MediaType.parse("application/json")))
                .build();
        // 3. Execute and handle response
        try (Response response = client.newCall(request).execute()) {
            if (!response.isSuccessful()) {
                System.err.println("❌ Call failed, HTTP status: " + response.code());
                return null;
            }
            String respBody = response.body() != null ? response.body().string() : "";
            ObjectNode respJson = (ObjectNode) mapper.readTree(respBody);
            String reply = respJson.path("choices").path(0).path("message").path("content").asText();
            int promptTokens = respJson.path("usage").path("prompt_tokens").asInt(0);
            int completionTokens = respJson.path("usage").path("completion_tokens").asInt(0);
            System.out.println("✅ Call succeeded!");
            System.out.println("📝 Response: " + reply);
            System.out.println("📊 Token usage – input: " + promptTokens + ", output: " + completionTokens);
            return reply;
        } catch (IOException e) {
            System.err.println("❌ Network or parsing error: " + e.getMessage());
            return null;
        }
    }
}

The client sends a POST request to {API_BASE_URL}/chat/completions with an OpenAI‑compatible JSON body, extracts choices[0].message.content as the answer, and reads the usage fields for token statistics.

Typical use cases

RAG knowledge‑base Q&A: convert documents to vectors, then use Qwen for semantic retrieval and answer generation.

AI coding assistant: integrate Qwen3.6 into tools such as OpenClaw, Cursor, or Claude Code for code completion, review, and unit‑test generation.

Multilingual customer‑service bot: leverage multilingual capability for automated support.

Structured data extraction: parse PDFs, emails, or reports to extract fields; NL2Repo benchmark shows strong performance.

AI agent development and testing: run agent workloads without token cost.

Key dates

Free unlimited token access ends on 30 June 2026 . After that, normal pricing applies.