Qwen3-Next-80B-A3B-Thinking is a Qwen3-Next model that supports only thinking mode. It is presented as the reasoning-focused variant of the Qwen3-Next line and is described as a model for complex reasoning tasks.
The model is associated with Alibaba’s Qwen team and is available through multiple host platforms, including Hugging Face and Alibaba Cloud-linked ecosystem pages.
| Developer | Alibaba Cloud |
|---|---|
| Model ID | alibaba/qwen3-next-80b-a3b-thinking |
| Release Date | September 29, 2025 |
| Context Window | 126,976 tokens (~127K) |
| Max Output Tokens | 81,920 tokens |
| Modalities | Text → Text |
| Capabilities | parallel tool calls, streaming, structured (JSON schema) output, tool/function calling, web search |
Thinking-only behavior
The model supports only thinking mode, and sources note that it outputs structured thinking traces by default.
Reasoning focus
It is described as optimized for complex reasoning tasks and hard multi-step problems.
Qwen3-Next family
It belongs to the Qwen3-Next series and is presented as part of the same architecture family as the broader Qwen3-Next line.
Alibaba/Qwen origin
The model is attributed to Alibaba’s Qwen team and Alibaba Cloud ecosystem coverage.
Availability in model hubs
The exact model id appears on Hugging Face and other model catalog pages, indicating public availability as a named model artifact.
Complex reasoning
Suitable for tasks described as complex reasoning, including multi-step analytical work.
Mathematical problem solving
Listed as ideal for transparent logic and problem solving tasks that benefit from visible thinking traces.
Code debugging and synthesis
Cited as useful for intricate code debugging and code synthesis tasks.
Agent planning
Presented as a fit for multi-step agentic planning workflows.
Qwen3-Next-80B-A3B-Base
The Alibaba Cloud blog identifies Qwen3-Next-80B-A3B-Base as the inaugural model in the architecture, while the Thinking variant is presented separately for reasoning tasks.
Qwen3-235B-A22B-Thinking-2507
Alibaba Cloud says Qwen3-Next-80B-A3B-Thinking approaches the performance of Qwen3-235B-A22B-Thinking-2507 on benchmarks.
Gemini-2.5-Flash-Thinking
Several model pages state that Qwen3-Next-80B-A3B-Thinking outperforms Gemini-2.5-Flash-Thinking on multiple benchmarks.
Sources: https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking · https://www.alibabacloud.com/blog/602536 · https://www.qubrid.com/models/qwen3-next-80b-a3b-thinking
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Qwen3-Next-80B-A3B-Thinking is documented as suited to highly complex reasoning tasks, and Alibaba says it excels in them. It is also documented as useful for tool calling capabilities.
It is described as specialized for complex problem-solving and deep reasoning, with a visible step-by-step reasoning process that makes it suitable for tasks requiring transparent logic. The documentation specifically calls out mathematical proofs as an example of such tasks.
It is also described as ideal for intricate code debugging. In the same documentation, it is presented as suitable for multi-step agent planning, where step-by-step reasoning is needed.
For broader application areas, it is documented as suitable for AI agent systems, chatbots, and RAG systems.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "alibaba/qwen3-next-80b-a3b-thinking", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "alibaba/qwen3-next-80b-a3b-thinking", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"alibaba/qwen3-next-80b-a3b-thinking","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
This model is thinking-mode only, with pricing listed separately for input and thinking output tokens, and Alibaba Cloud notes the displayed prices are original API-call pricing excluding limited-time promotions.
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 11.2 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | September 12, 2026 |
| Coding | 17.4 | Composite score across standardised coding evaluations, measured independently by Artificial Analysis | Source | September 12, 2026 |
| Math | 84.3 | Composite score across standardised mathematics evaluations, measured independently by Artificial Analysis | Source | September 12, 2026 |
| AIME 2025 | 87.8 | Competition mathematics (AIME), 2025 | Source | September 22, 2026 |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Qwen3 Next 80B A3B Thinking This page | Agentic workflows and structured output | |||
| Coding + agents | ||||
| Coding + agents | ||||
| Coding + agents |
Qwen3 Next 80B A3B Thinking has a 126,976 tokens context window and can return up to 81,920 tokens.
Qwen3 Next 80B A3B Thinking takes text as input and returns text.
Use alibaba/qwen3-next-80b-a3b-thinking as the model id. Requests go to https://api.aimlapi.com/v1/chat/completions.
Qwen3 Next 80B A3B Thinking became available on September 30, 2025.
Qwen3 Next 80B A3B Thinking is priced at input $0.195 / 1M tokens, output $1.56 / 1M tokens.
Yes, Qwen3 Next 80B A3B Thinking can stream responses as they are generated.
Qwen3 Next 80B A3B Thinking was built by Alibaba Cloud.
It is designed as a multilingual reasoning model with chain-of-thought capabilities, suited for complex reasoning tasks.
Yes, it supports tool use, including parallel tool calls and structured output.
