Qwen3.6 Plus
Qwen3.6 Plus by Alibaba is a proprietary multimodal agent model with 1M-token context, preserve_thinking API, and agentic coding for repository-level tasks.
Model Overview
Capabilities, design details, and architectural traits
Qwen3.6 Plus - Alibaba's "Towards Real World Agents"
Qwen3.6 Plus is a proprietary, hosted model from Alibaba available via Alibaba Cloud Model Studio. Its official framing is "Towards Real World Agents" - it is designed around a "capability loop": perceiving, reasoning, and acting within a single workflow rather than responding passively. It accepts text and image/video input and produces text output. Alibaba describes it as an "all-rounder" that integrates deep logical reasoning, extensive contextual memory, and precise tool execution within a unified agent session.
What Defines Qwen3.6 Plus Within the 3.6 Family
| Trait | Detail |
|---|---|
preserve_thinking API parameter | Introduced with this release; carries `` reasoning blocks across all preceding turns; disabled by default; Alibaba explicitly recommends it for agentic tasks to maintain decision consistency and reduce redundant reasoning |
| 1M-token context window by default | Designed for repository-level engineering; the full 1M is available without any special configuration |
| Multimodal across three documented dimensions | Advanced reasoning (document understanding, video reasoning, visual coding), real-world applicability (retail intelligence, fine-grained visual perception), and complex task execution |
| Six named third-party coding framework integrations | Officially documented: OpenClaw, Claude Code, Qwen Code, Kilo Code, Cline, and OpenCode |
| Proprietary API-only, closed weights | Contrast with the open-weight Qwen3.6-35B-A3B (Flash tier) released under Apache 2.0 the same month |
Agentic Coding as the Primary Design Goal
The official blog explicitly states that Qwen3.6 Plus autonomously plans, tests, and iterates on code across the full execution loop - from objective breakdown to final refinement - targeting both frontend web development and complex repository-level engineering. The preserve_thinking parameter is the specific mechanism introduced to maintain reasoning context across the multiple tool calls this loop requires.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Alibaba | Anthropic | Anthropic |
| Release Date | April 2, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 262K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.50 Best Input Pricing | $4 | $2 |
| Output Pricing | $3 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | textimagevideo | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 27.0 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 54.5 | - | - |
| Agentic Index | 29.0 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Alibaba
Other models by Alibaba
Qwen3.8 Max
Qwen3.8 2.4T A95B
Qwen3.8-Flash-Next
Qwen3.8 27B
Top AI Models
Leading alternatives by intelligence score
