Large Language Models

Qwen3.6

by Alibaba Cloud · Freemium

Alibaba Cloud's Qwen3.6 family — three models covering frontier coding (Max-Preview), multimodal enterprise (Plus), and open-weight self-hosting (35B-A3B). Tops 6 coding benchmarks at $1.30/M input.

Visit Website ↗

4.0

Avg Score

from 1 reviews

1 Articles

4 Compared

0 Prompts

📋 Editorial Review

In-Depth Editorial ★ Editor's Pick

Qwen3.6 Review 2026: Alibaba's Best Model Tops 6 Coding Benchmarks

Qwen3.6-Max-Preview leads SWE-bench Pro, Terminal-Bench 2.0, and four more coding benchmarks at $1.30/M input — cheaper than GPT-5.4 and Claude Opus 4.7 at the same capability tier. Three models, three use cases, one family.

✎ pat bob 🕑 8 min read 👁 3,244 views

9.0

Score

Read Review →

About Qwen3.6

Qwen3.6 is Alibaba Cloud's April 2026 model family consisting of three tiers. Qwen3.6-Max-Preview (released April 20) is a closed-weights MoE flagship with 35B total / 3B activated parameters, 256K context, leading SWE-bench Pro, Terminal-Bench 2.0, QwenWebBench, SkillsBench, and SciCode. Scores 52 on the Artificial Analysis Intelligence Index — highest of any Chinese-origin model. Features preserve_thinking for persistent reasoning traces across multi-turn agentic workflows. Priced at $1.30/M input · $7.80/M output. Qwen3.6-Plus (released April 2) is the multimodal enterprise tier with a 1M-token context window supporting text, image, and video input at ~$0.40/M input. Qwen3.6-35B-A3B is the open-weight variant under Apache 2.0, scoring 73.4% on SWE-bench Verified, runnable on consumer hardware via Ollama or vLLM. All models use an OpenAI-compatible API.

✅ Key Features

Qwen3.6-Max-Preview: #1 SWE-bench Pro, 256K context, preserve_thinking, $1.30/M input, Qwen3.6-Plus: 1M token multimodal context (text/image/video), ~$0.40/M input, Qwen3.6-35B-A3B: Apache 2.0 open weights, 73.4% SWE-bench Verified, self-hostable via Ollama/vLLM, OpenAI-compatible API for all three models, 1M free tokens per model for new users (Singapore region)

🧩 Specs

Platforms

WebAPISelf-hosted

Integrations

OllamavLLMSGLangOpenRouterFireworksAlibaba Cloud BailianDingTalkWukong

🎯 Best For & Tags

Best For

Agentic coding pipelinesSWE-bench class software engineering tasksFront-end code generationLong-document and multimodal enterprise workflowsSelf-hosted private inferenceCI/CD repair automation

⭐ User Reviews

Write a Review →

★★★★☆

4/5 05 May 2026

#1 on SWE-bench Pro at $1.30/M — Alibaba just changed the coding AI price war

“Qwen3.6-Max-Preview tops six coding benchmarks at a price point below GPT-5.4 and Claude Opus 4.7. preserve_thinking for agentic workflows is the sleeper feature. Open weights on the 35B-A3B under Apache 2.0 keeps the ecosystem honest.”

By AITools

✓ Pros

Tops SWE-bench Pro and 5 other coding benchmarks at $1.30/M input. preserve_thinking for persistent multi-turn reasoning in agentic workflows. OpenAI-compatible API — zero integration changes. Apache 2.0 open-weight 35B-A3B self-hostable on consumer hardware. 1M free tokens per model for new users.

✗ Cons

Text-only flagship — no image or video. Chinese data governance laws apply to API tiers. No open-source release date for the flagship. Free OpenRouter tier collects prompts for training.