Spaces:
Running
A newer version of the Gradio SDK is available: 6.18.0
title: QWEN 3.5 Multichat
emoji: 🚀
colorFrom: purple
colorTo: purple
sdk: gradio
sdk_version: 6.8.0
app_file: app.py
pinned: false
short_description: Qwen3.5 official collection — 3 models
hf_oauth: true
✦ Qwen3.5 MultiChat — Multi-Model AI Chat with Thinking Mode & Vision
The world's most capable open-source LLM collection in one interface. Switch between 122B, 27B, and 35B models instantly. Think deeper. See further. Speak any language.
What is Qwen3.5 MultiChat?
Qwen3.5 MultiChat is a production-grade AI chat interface powered by Alibaba's official Qwen3.5 model collection. It lets you seamlessly switch between three state-of-the-art models — each optimized for different tasks — with real-time streaming, chain-of-thought reasoning, image understanding, and support for 201 languages.
No API key needed. Log in with your Hugging Face account and start chatting in seconds.
Models
| Model | Type | Benchmark | Best For |
|---|---|---|---|
| Qwen3.5-122B-A10B | MoE · 10B active | BFCL 72.2 · GPQA 86.6 · SWE 72.0 | Complex reasoning, agents, math |
| Qwen3.5-27B | Dense · 27B active | IFEval 95.0 · SWE 72.4 · PolyMATH 71.2 | Creative writing, coding, translation |
| Qwen3.5-35B-A3B | MoE · 3B active | TAU2 81.2 · MMLU-Pro 85.3 | Fast responses, everyday tasks |
Features
- 🧠 Thinking Mode — Exposes the model's internal reasoning chain before answering. Best for math, logic, and complex multi-step problems.
- 👁️ Vision — Upload any image and ask questions. All three models support multimodal input.
- 🌏 201 Languages — Native multilingual support including Korean, Japanese, Arabic, and more.
- 📄 262K Context Window — Feed entire documents, codebases, or long conversations without chunking.
- ⚡ Real-Time Streaming — Token-by-token SSE streaming with live rendering.
- 🔐 HF OAuth — One-click sign-in with your Hugging Face account.
- 🎛️ Full Parameter Control — Adjust temperature, top-p, max tokens, and system prompt per session.
Frequently Asked Questions
What makes Qwen3.5 different from other open-source LLMs? Qwen3.5 achieves GPT-4-class performance on most benchmarks while being fully open-weight. The 122B MoE model uses only 10B active parameters per token, making it dramatically more efficient than comparable dense models.
What is Thinking Mode? Thinking Mode instructs the model to generate an internal chain-of-thought before producing its final answer. This significantly improves accuracy on mathematical reasoning, logical inference, and multi-step problem solving — similar to OpenAI's o1/o3 reasoning models.
Is this free to use? Yes. Sign in with a Hugging Face account to use the Inference API at no cost, subject to standard HF rate limits.
Which model should I choose? Use 122B-A10B for deep reasoning and agentic tasks. Use 27B for writing, translation, and instruction following. Use 35B-A3B when you need fast responses with low latency.
Can it analyze images? All three models support vision input. Upload a screenshot, diagram, photo, or document scan and ask any question about it.
What languages does it support? Qwen3.5 was trained on 201 languages. Performance is especially strong in English, Chinese, Korean, Japanese, French, German, Spanish, and Arabic.
How long can my input be? Up to 262,144 tokens (~200,000 words). This is enough to fit most research papers, codebases, or book chapters in a single prompt.
Tech Stack
Qwen3.5 · Gradio 6 · FastAPI · Hugging Face Inference API · HF OAuth 2.0 · Server-Sent Events · Python 3.13
Hashtags
#Qwen3 #Qwen35 #OpenSourceAI #LLM #AIChat #MultimodalAI #ThinkingAI #ReasoningAI #VisionLanguageModel #FreeAI #AIAssistant #ChatAI #HuggingFace #GenerativeAI #LargeLanguageModel #AIModel #SOTA #MixtureOfExperts #MoE #ChainOfThought #CoT #AIReasoning #NaturalLanguageProcessing #NLP #DeepLearning #OpenWeightModel #GINIGENAI #AIStartup #ProductionAI #StreamingAI #MultilingualAI