Skip to main content
Qubrid AI
Alibaba Cloud · Chat / LLM · 80B Parameters (3.9B Active) · 256K ContextQubrid Playground License HuggingFaceStreaming Reasoning Chain-of-Thought Long Context Code Agentic Planning

Overview

Qwen3 Next 80B A3B Thinking is a next-generation foundation model from Alibaba’s Qwen team featuring a revolutionary Hybrid Attention mechanism (Gated DeltaNet + Gated Attention) with High-Sparsity MoE architecture. With 80B total parameters and only 3.9B active per token, it delivers 10x higher throughput than Qwen3-32B on long contexts while outperforming Gemini-2.5-Flash-Thinking on multiple benchmarks. Designed exclusively for deep reasoning tasks, it operates in thinking-only mode — surfacing full chain-of-thought traces before every response. Served instantly via the Qubrid AI Serverless API.
🧠 10x throughput vs Qwen3-32B. Outperforms Gemini-2.5-Flash-Thinking. 3.9B active parameters. Deploy on Qubrid AI — no infrastructure required.

Model Specifications

FieldDetails
Model IDQwen/Qwen3-Next-80B-A3B-Thinking
ProviderAlibaba Cloud (Qwen Team)
KindChat / LLM
ArchitectureHybrid Transformer-Mamba (Gated DeltaNet + Gated Attention) with High-Sparsity MoE and Multi-Token Prediction (MTP)
Parameters80B total (3.9B active per token)
Context Length256,000 Tokens
MoENo
Release DateSeptember 2025
LicenseApache 2.0
Training DataLarge-scale multilingual pretraining dataset, fine-tuned with GSPO for thinking
Function CallingNot Supported
Image SupportN/A
Serverless APIAvailable
Fine-tuningComing Soon
On-demandComing Soon
State🟢 Ready

Pricing

💳 Access via the Qubrid AI Serverless API with pay-per-token pricing. No infrastructure management required.
Token TypePrice per 1M Tokens
Input Tokens$0.20
Output Tokens$1.80

Quickstart

Prerequisites

  1. Create a free account at platform.qubrid.com
  2. Generate your API key from the API Keys section
  3. Replace QUBRID_API_KEY in the code below with your actual key
💡 Thinking mode: This model always produces chain-of-thought reasoning traces before its final answer. Plan for higher output token counts on complex tasks — use max_tokens accordingly.

Python

from openai import OpenAI

# Initialize the OpenAI client with Qubrid base URL
client = OpenAI(
    base_url="https://platform.qubrid.com/v1",
    api_key="QUBRID_API_KEY",
)

# Create a streaming chat completion
stream = client.chat.completions.create(
    model="Qwen/Qwen3-Next-80B-A3B-Thinking",
    messages=[
      {
        "role": "user",
        "content": "Explain quantum computing in simple terms"
      }
    ],
    max_tokens=8192,
    temperature=0.6,
    top_p=0.95,
    stream=True
)

# If stream = False comment this out
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
print("\n")

# If stream = True comment this out
print(stream.choices[0].message.content)

JavaScript

import OpenAI from "openai";

// Initialize the OpenAI client with Qubrid base URL
const client = new OpenAI({
  baseURL: "https://platform.qubrid.com/v1",
  apiKey: "QUBRID_API_KEY",
});

// Create a streaming chat completion
const stream = await client.chat.completions.create({
  model: "Qwen/Qwen3-Next-80B-A3B-Thinking",
  messages: [
    {
      role: "user",
      content: "Explain quantum computing in simple terms",
    },
  ],
  max_tokens: 8192,
  temperature: 0.6,
  top_p: 0.95,
  stream: true,
});

// If stream = false comment this out
for await (const chunk of stream) {
  if (chunk.choices[0]?.delta?.content) {
    process.stdout.write(chunk.choices[0].delta.content);
  }
}
console.log("\n");

// If stream = true comment this out
console.log(stream.choices[0].message.content);

Go

package main

import (
	"bufio"
	"bytes"
	"encoding/json"
	"fmt"
	"net/http"
)

func main() {
	url := "https://platform.qubrid.com/v1/chat/completions"

	data := map[string]interface{}{
		"model": "Qwen/Qwen3-Next-80B-A3B-Thinking",
		"messages": []map[string]string{
			{
				"role":    "user",
				"content": "Explain quantum computing in simple terms",
			},
		},
		"temperature": 0.6,
		"max_tokens":  8192,
		"stream":      true,
		"top_p":       0.95,
	}

	jsonData, _ := json.Marshal(data)
	req, _ := http.NewRequest("POST", url, bytes.NewBuffer(jsonData))
	req.Header.Set("Authorization", "Bearer QUBRID_API_KEY")
	req.Header.Set("Content-Type", "application/json")

	client := &http.Client{}
	res, _ := client.Do(req)
	defer res.Body.Close()

	scanner := bufio.NewScanner(res.Body)
	for scanner.Scan() {
		line := scanner.Text()
		if line != "" {
			fmt.Println(line)
		}
	}
}

cURL

curl -X POST "https://platform.qubrid.com/v1/chat/completions" \
  -H "Authorization: Bearer QUBRID_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "Qwen/Qwen3-Next-80B-A3B-Thinking",
  "messages": [
    {
      "role": "user",
      "content": "Explain quantum computing in simple terms"
    }
  ],
  "temperature": 0.6,
  "max_tokens": 8192,
  "stream": true,
  "top_p": 0.95
}'

Live Example

Prompt: Prove that the square root of 2 is irrational
Response:
<think>
This is a classic proof by contradiction. I'll use the standard approach:
assume √2 is rational, express it as a fraction in lowest terms, and
derive a contradiction by showing both numerator and denominator must
be even — contradicting the assumption that the fraction is in lowest terms.
</think>

Proof: √2 is irrational

Assume for contradiction that √2 is rational.
Then √2 = p/q where p, q are integers with no common factors (lowest terms).

Step 1: Square both sides
  2 = p²/q²
  → p² = 2q²

Step 2: p² is even, therefore p is even
  Since p² = 2q², p² is divisible by 2.
  If p² is even, then p must be even (odd × odd = odd, so odd² is always odd).
  Let p = 2k for some integer k.

Step 3: Substitute back
  (2k)² = 2q²
  4k²   = 2q²
  q²    = 2k²

Step 4: q² is even, therefore q is even
  By the same logic as Step 2, q must also be even.

Step 5: Contradiction
  Both p and q are even — meaning they share a common factor of 2.
  This contradicts our assumption that p/q is in lowest terms.

Conclusion:
  Our assumption that √2 is rational leads to a contradiction.
  Therefore √2 is irrational. ∎
Try it yourself in the Qubrid AI Playground →

Playground Features

The Qubrid AI Playground lets you interact with Qwen3 Next 80B Thinking directly in your browser — no setup, no code, no cost to explore.

🧠 System Prompt

Define the model’s reasoning approach, domain focus, and output format before the conversation begins. Particularly powerful for mathematical proofs, agentic planning, and structured analytical tasks.
Example: "You are an expert mathematical reasoning assistant. Work through
every problem step by step, state all assumptions explicitly, and present
your final answer in a clearly marked conclusion block."
Set your system prompt once in the Qubrid Playground and it applies across every turn of the conversation.

🎯 Few-Shot Examples

Guide the model’s reasoning style and output structure with concrete examples — no fine-tuning, no retraining required.
User InputAssistant Response
Is 97 a prime number?Yes. 97 is prime. Check divisibility by all primes ≤ √97 ≈ 9.8: not divisible by 2, 3, 5, or 7. Therefore 97 has no factors other than 1 and itself.
Write a Python function to check if a string is a palindromedef is_palindrome(s: str) -> bool: s = s.lower().replace(" ", "") return s == s[::-1]
💡 Stack multiple few-shot examples in the Qubrid Playground to establish your preferred reasoning format and output structure — no fine-tuning required.

Inference Parameters

ParameterTypeDefaultDescription
StreamingbooleantrueEnable streaming responses for real-time output
Temperaturenumber0.6Controls randomness. Lower values recommended for reasoning tasks
Max Tokensnumber8192Maximum number of tokens to generate
Top Pnumber0.95Nucleus sampling parameter

Use Cases

  1. Complex multi-step reasoning
  2. Mathematical proofs
  3. Code synthesis
  4. Logical analysis
  5. Agentic planning
  6. Long-context document analysis

Strengths & Limitations

StrengthsLimitations
Hybrid Attention (Gated DeltaNet + Gated Attention) for efficient long-context processingThinking mode only — no fast non-thinking mode available
10x throughput vs Qwen3-32B on 32K+ contextsLonger thinking traces increase latency on complex tasks
Only 3.9B active parameters from 80B total (efficient inference)New architecture with limited community tooling support
Native 256K context windowFunction calling not supported
Outperforms Gemini-2.5-Flash-Thinking on multiple benchmarks
Apache 2.0 — fully open-source with commercial use

Why Qubrid AI?

  • 🚀 No infrastructure setup — 80B MoE served serverlessly, pay only for what you use
  • 🔁 OpenAI-compatible — drop-in replacement using the same SDK, just swap the base URL
  • 🧠 Reasoning at scale — Qwen3 Next’s 10x throughput advantage is fully realized on Qubrid’s low-latency infrastructure
  • 🧪 Built-in Playground — prototype with system prompts and few-shot examples instantly at platform.qubrid.com
  • 📊 Full observability — API logs and usage tracking built into the Qubrid dashboard
  • 🌐 Multi-language support — Python, JavaScript, Go, cURL out of the box

Resources


Built with ❤️ by Qubrid AI

Frontier models. Serverless infrastructure. Zero friction.