Qwen3

Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

Qwen3

Qwen3 is the third-generation text model family from Alibaba Cloud's Qwen team, released on April 28, 2025. It covers six dense sizes (0.6B to 32B) and two MoE variants, all trained on approximately 36 trillion tokens across 119 languages. A key design choice is a unified thinking and non-thinking mode in every model, so developers can choose between step-by-step reasoning and fast single-pass responses without switching models.

General
Release date	28 Apr 2025
Developer	Qwen / Alibaba Cloud
Type	Open-weight text LLM family
License	Apache 2.0
GitHub	QwenLM/Qwen3
Hugging Face	huggingface.co/Qwen
Documentation	qwenlm.github.io/blog/qwen3

Core Features

Thinking/non-thinking mode: every model supports both step-by-step chain-of-thought reasoning and direct response generation within a single checkpoint.
Thinking budget: developers can set a token budget for the reasoning phase, allowing inference cost to be tuned per request.
Long context: models at 4B and above support 131,072-token context windows; 0.6B and 1.7B support 32,768 tokens.
Multilingual: pretrained on 119 languages and dialects.
Apache 2.0: all weights are released for commercial use, fine-tuning, and redistribution.

Model Variants

Variant	Total Params	Active Params	Context	Best for
Qwen3-0.6B	0.6B	0.6B	32K	Edge and on-device
Qwen3-1.7B	1.7B	1.7B	32K	Lightweight inference
Qwen3-4B	4B	4B	128K	Balanced performance
Qwen3-8B	8B	8B	128K	General tasks
Qwen3-14B	14B	14B	128K	Higher accuracy
Qwen3-32B	32B	32B	128K	Strong reasoning
Qwen3-30B-A3B	30B	3B	128K	Efficient MoE
Qwen3-235B-A22B	235B	22B	128K	Flagship MoE

Benchmarks

The flagship Qwen3-235B-A22B model scores:

AIME '24: 85.7
AIME '25: 81.5
LiveCodeBench v5: 70.7
BFCL v3: 70.8

Tools and Resources

Qwen3 Blog Post: official release notes and technical details.
Hugging Face (Qwen): download model weights and GGUF files.
Technical Report: arXiv paper covering architecture, training, and evaluation.
Ollama: run Qwen3 models locally.
Qwen API Platform: access models via OpenAI-compatible API.

Ecosystem and Integrations

Available on Hugging Face Hub in both standard and GGUF formats.
Accessible via Alibaba Cloud DashScope using an OpenAI-compatible endpoint.
Supported by Ollama, LM Studio, and major inference frameworks including vLLM and llama.cpp.
All sizes available for fine-tuning using standard supervised fine-tuning and RL pipelines.

Qwen3 weights are available immediately on Hugging Face. To access via API, generate a key on the Qwen API Platform and follow the Model Studio documentation.

Edit on GitHub

Qwen Qwen3 AI technology Hackathon projects

Discover innovative solutions crafted with Qwen Qwen3 AI technology, developed by our community members during our engaging hackathons.

SmartBandit Router

BanditRoute treats each AI model as an "arm" in a multi-armed bandit problem. Using Thompson Sampling conditioned on task type (code, math, QA, creative), it learns over time which model to send each query to balancing correctness against token cost and latency via a custom reward function, rather than optimizing for accuracy alone. No routing rule was ever hand-written; behaviors like sending math questions to the strongest model and easy QA to a cheap one emerged entirely from feedback. The project is validated two ways: a full simulation (2000 queries, 4 mock model profiles) that shows clean convergence and ~93% cost savings, and a live integration that runs the same bandit logic against real free-tier models via the OpenRouter API proving the approach works beyond simulation, not just in theory.

AI Classroom Edge Intelligence

AI Classroom Edge Intelligence is a privacy-first classroom AI platform built for schools that need useful AI without sending every piece of student information to the cloud. The platform evaluates each task by privacy level, connectivity, and complexity, then routes it to Offline Edge Mode, a Local Classroom Server, or Fireworks AI Cloud Assist. Sensitive or restricted information stays local. Eligible anonymized, high-complexity tasks are sent through a secure Express backend to Fireworks Serverless using Qwen3.7 Plus. The project includes an AMD Model Router, Edge Runtime Monitor, Rural Connectivity Simulator, Privacy and Local Data Ownership Console, Classroom Digital Twin, and teacher approval workflow. Live results display the provider, model, route, privacy classification, latency, safety note, and AI response. API keys remain server-side, and the privacy guard blocks sensitive requests from cloud inference. The project was inspired by rural schools where connectivity can be unreliable and student privacy is critical. Instead of acting as a simple chatbot, it serves as an intelligent routing and decision-support system. Teachers review, edit, approve, or reject recommendations before instructional actions are recorded. The current implementation includes a working browser interface, real backend routing, live Fireworks Serverless integration, server-side key protection, and Docker containerization. Local AMD AI PC inference, GPU/NPU acceleration, device telemetry, and production synchronization are clearly identified as future work. The long-term vision is a school-owned AI platform combining local intelligence, optional cloud reasoning, persistent classroom evidence, and teacher oversight for rural and underserved communities.

Claption

Claption is a containerized video captioning app built for the AMD Developer Hackathon ACT II Video Captioning track. It processes short video clips by sampling frames with FFmpeg, generating a grounded visual fact sheet with Fireworks AI, and then rewriting that factual summary into four required caption styles: formal, sarcastic, humorous-tech, and humorous-non-tech. To improve leaderboard performance, Claption separates factual understanding from tone generation, then uses an internal LLM judge loop to score captions for accuracy, tone match, humor quality, and hallucination risk. Weak captions are automatically repaired before final export. The project includes a Next.js demo app, batch CLI, Docker container, sample outputs, and judge-ready setup instructions.

DispatchAI is an autonomous emergency grid

During mass-casualty events or natural disasters, human dispatchers become severely overwhelmed by chaotic, unstructured reports . Crucial seconds are lost parsing text, mapping locations, and finding the right medic, which causes delayed response times when seconds mean the difference between life and death dispatchAI is a next-generation autonomous emergency grid powered by AMD designed to eliminate this human bottleneck . It provides autonomous coordination at scale by using Natural Language Processing to instantly ingest unstructured emergency reports via Telegram webhooks . Our system scales to handle 10 simultaneous mass-casualty incidents in under 10 seconds, successfully reducing average dispatch latency by 90% . The architecture operates as an Event-Driven Asynchronous Pipeline, utilizing a FastAPI backend and a tactical React/Vite dashboard . At its core is the "AMD Brain," which is powered by AMD Instinct™ MI300X Accelerators running the Qwen2.5-14B-Instruct model . We chose this model because it offers the perfect balance of deep reasoning for complex medical routing and high inference speed directly on AMD hardware . The LLM is rigorously prompt-engineered to align with human dispatcher logic, matching the exact skills needed with the closest available volunteer . Furthermore, the system includes dynamic "Follow-Up" logic via Telegram to clarify vague emergencies, and features a 15-second graceful fallback to pure proximity math to ensure enterprise reliability without AI hallucinations . DispatchAI ensures the right help arrives exactly when it is needed

VigiAI focused on the alert feature

VigiAI is an intelligent real-time surveillance system for detecting traffic accidents and anomalies. Running on an AMD MI300X accelerator, its hybrid architecture splits tasks between local processing—using Qwen-VL 7B, YOLO v8x, optical flow, and pose estimation—and cloud analytics via DeepSeek-V4 Flash through the Fireworks AI API. To handle chaotic real-world events, it was trained on the complete UCF-Crimes dataset (over 100 GB of footage). The system processes video through a structured 7-phase pipeline: Event Detection: Identifies anomalies instantly. Localization: Pins down spatial bounding boxes and timestamps. Scene Analysis: Evaluates weather, road, and lighting conditions. Causal Inference: Infers accident causes and secondary risks. Object Tracking: Tracks and counts vehicles or pedestrians. Speed Estimation: Measures velocity using pixel-to-meter calibration. Narrative Generation: Compiles an executive summary and detailed report. VigiAI detects 14 anomaly classes (accidents, theft, fights, fires, etc.) and transmits instant alerts via WebSockets over a public Cloudflare Tunnel. The web dashboard features live camera/webcam feeds, a GPS map, historical logs, and local vs. cloud performance metrics.

Visiolab

VisioLab is a state-of-the-art educational ecosystem designed to revolutionize how students interact with and understand complex STEM concepts (Science, Technology, Engineering, and Mathematics). By leveraging advanced Artificial Intelligence, VisioLab bridges the gap between theoretical textbook knowledge and practical application through real-time, interactive simulations. The platform operates on a synchronized dual-portal architecture: The Admin Portal acts as a powerful creative engine for educators. Teachers can seamlessly generate brand-new, fully functional simulations simply by providing a text prompt or reference material. The integrated AI automatically writes the underlying code to visualize abstract concepts—ranging from cellular biology to harmonic physics—giving educators a frictionless way to enrich their curriculum. The Student Portal offers a premium, beautifully designed learning environment where students can actively manipulate variables, observe live metric updates, and watch graphs respond in real-time. To ensure deep comprehension, each interactive model is paired with dynamically generated AI Study Guides and a context-aware conversational AI Tutor capable of providing step-by-step guidance. Built entirely with modern web technologies including React, Tailwind CSS, and Supabase, VisioLab delivers a highly secure, responsive, and deeply engaging educational experience that empowers the curious minds of tomorrow.