# Qwen3.6-Plus

> Alibaba's 1M-context multimodal agent model for real-world coding

- Maker: Alibaba (Qwen)
- Line: Qwen-Plus (multimodal agent)
- Canonical page: https://ai-tldr.dev/models/qwen3-6-plus/
- Updated: 2026-06-18

## In short

Qwen3.6-Plus is the proprietary flagship of Alibaba's Qwen line, released on April 2, 2026 by the Qwen team. It has a 1M-token context window and costs $0.50 per million input tokens and $3.00 per million output tokens. You can use it through Alibaba Cloud Model Studio (Bailian) and OpenRouter.

## Facts

| Fact | Value | Source |
| --- | --- | --- |
| Context window | 1M (max output 65,536 tokens) | [1] |
| Input price | $0.50 per 1M tokens | [2] |
| Output price | $3.00 per 1M tokens | [2] |
| Licence | Proprietary (API-only) · API only | [1] |
| Released | April 2026 | [3] |
| Parameters | Undisclosed | [1] |
| Knowledge cutoff | Not disclosed | [1] |

## Sources

1. [Alibaba Cloud Model Studio — Qwen API reference](https://www.alibabacloud.com/help/en/model-studio/qwen-api-reference/)
2. [Qwen3.6-Plus benchmarks & pricing (LLM-Stats)](https://llm-stats.com/models/qwen3.6-plus)
3. [Alibaba press release: Qwen3.6-Plus for agentic AI](https://treasurytoday.com/press-releases/press-release-alibaba-unveils-qwen-3-6-plus-to-accelerate-agentic-ai-deployment-for-enterprises-and-alibabas-ai-applications/)

## Overview

Qwen3.6-Plus is the proprietary flagship of Alibaba's Qwen line, released on April 2, 2026 by the Qwen team. It is positioned as a multimodal agent model: it accepts text, images, UI screenshots, and video, runs a 1-million-token context window by default, and can autonomously plan, write, test, and iterate on code within a single workflow. Alibaba pitches it for repository-level engineering and for turning designs, wireframes, and prototypes directly into working front-end code.

Architecturally, Qwen3.6-Plus uses a hybrid design that pairs linear attention with a sparse Mixture-of-Experts, which Alibaba uses to keep inference efficient at the 1M-token scale. Reasoning (chain-of-thought) is always on, and the model ships native function calling and tool use. Alibaba has not published parameter counts or a knowledge-cutoff date for the Plus tier.

Qwen3.6-Plus itself is proprietary and API-only, served through Alibaba Cloud Model Studio (Bailian) and Qwen Chat, with OpenAI- and Anthropic-compatible endpoints so it drops into tools like Claude Code, Cline, and OpenClaw. Note that it is distinct from the separately released open-weight Qwen3.6 models (Qwen3.6-27B dense and Qwen3.6-35B-A3B MoE), which ship under Apache 2.0 — the Plus weights are not public.

## Benchmarks

- SWE-bench Verified: 78.8% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- Terminal-Bench 2.0: 61.6% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- OSWorld-Verified: 62.5% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- MCPMark: 48.2% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- DeepPlanning: 41.5% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- OmniDocBench v1.5: 91.2% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- RealWorldQA: 85.4% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- MMMU: 86% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))
- GPQA: 90.4% ([source](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide))

## Pricing

- Input: $0.50 per 1M tokens
- Output: $3.00 per 1M tokens

Standard API rate. Model Studio applies tiered, prompt-length-based billing, and some channels (e.g. OpenRouter) list discounted rates around $0.325 input / $1.95 output. Confirm the live rate for your region.

Source: https://llm-stats.com/models/qwen3.6-plus

## Strengths

- 1-million-token native context window for repository-scale and long-document tasks
- Agentic coding that leads Claude 4.5 Opus on Terminal-Bench 2.0 (61.6 vs 59.3)
- Best-in-class document parsing — 91.2 on OmniDocBench v1.5
- Native multimodal input: text, images, UI screenshots, and long-form video
- Always-on chain-of-thought reasoning with native function calling and tool use
- Drop-in via OpenAI- and Anthropic-compatible APIs (Claude Code, Cline, OpenClaw)
- Aggressive price/performance versus Western frontier models

## Best for

- Autonomous, repository-level coding agents that plan, write, test, and iterate
- Front-end and visual coding: turning screenshots, wireframes, or mockups into working UI code
- High-density document parsing and extraction (forms, tables, scanned PDFs as images)
- Long-context analysis across large codebases or document sets up to 1M tokens
- Multimodal reasoning over images and long-form video
- GUI / computer-use automation tasks (OSWorld-style desktop agents)

## How to access

- [Alibaba Cloud Model Studio (Bailian)](https://www.alibabacloud.com/help/en/model-studio/qwen-api-reference/) — model id `qwen3.6-plus`
- [OpenRouter](https://openrouter.ai/qwen/qwen3.6-plus) — model id `qwen/qwen3.6-plus`

## Qwen-Plus (multimodal agent) versions

- [Qwen3.7-Plus](https://ai-tldr.dev/models/qwen3-7-plus/) (2 June 2026)
- Qwen3.6-Plus (April 2026) — this page
- [Qwen3.5-Plus](https://ai-tldr.dev/models/qwen3-5-plus/) (16 February 2026)

## FAQ

### Is Qwen3.6-Plus open source?

No. The Qwen3.6-Plus flagship is proprietary and API-only, served through Alibaba Cloud Model Studio (Bailian) and Qwen Chat. Alibaba does separately release open-weight Qwen3.6 models — Qwen3.6-27B (dense) and Qwen3.6-35B-A3B (MoE) — under the Apache 2.0 license, but those are different models from the Plus tier and the Plus weights are not published.

### What is the context window and max output of Qwen3.6-Plus?

Qwen3.6-Plus runs a 1-million-token context window by default and can generate up to 65,536 output tokens per response.

### How does Qwen3.6-Plus compare to Claude on coding?

On agentic coding it edges ahead of Claude 4.5 Opus on Terminal-Bench 2.0 (61.6 vs 59.3), and on SWE-bench Verified it scores 78.8 versus Claude 4.5 Opus's reported 80.9 — competitive but slightly behind on that specific benchmark.

### What can Qwen3.6-Plus take as input?

It is multimodal: it accepts text, images (including UI screenshots, wireframes, and high-density documents), and long-form video. Document parsing is a headline strength, with a 91.2 score on OmniDocBench v1.5.

## Links

- [Qwen3.6-Plus: Towards Real World Agents (official blog)](https://qwen.ai/blog?id=qwen3.6)
- [Alibaba press release: Qwen3.6-Plus for agentic AI](https://treasurytoday.com/press-releases/press-release-alibaba-unveils-qwen-3-6-plus-to-accelerate-agentic-ai-deployment-for-enterprises-and-alibabas-ai-applications/)
- [Alibaba Cloud Model Studio — Qwen API reference](https://www.alibabacloud.com/help/en/model-studio/qwen-api-reference/)
- [Qwen3.6-Plus on OpenRouter (pricing & API)](https://openrouter.ai/qwen/qwen3.6-plus)
- [Qwen3.6-Plus benchmarks & pricing (LLM-Stats)](https://llm-stats.com/models/qwen3.6-plus)
- [Qwen 3.6 Plus Complete Guide (Serenities AI)](https://serenitiesai.com/articles/qwen-3-6-plus-complete-guide)
- [Constellation Research: Qwen launches flagship Qwen3.6-Plus](https://www.constellationr.com/insights/news/alibabas-qwen-launches-new-flagship-llm-qwen-36-plus)
- [QwenLM/Qwen3.6 GitHub (open-weight 3.6 models)](https://github.com/QwenLM/Qwen3.6)

Source: AI/TLDR LLM registry, https://ai-tldr.dev/models/qwen3-6-plus/