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.

TradeNexus AI — Autonomous Trade Intelligence

TradeNexus AI solves a critical gap: 400 million SMEs that trade globally have no access to enterprise-grade trade intelligence. They make multi-million dollar decisions using Google Search and gut feeling. Our platform deploys 11 autonomous AI agents across 4 modules: MODULE A — SupplierPulse: Real-time supplier risk scoring (0-100) using news intelligence, PDF financial document analysis, factory image inspection, and free OSINT cybersecurity scanning — SSL certificates, exposed database ports, and data breach history. Complete risk picture in 60 seconds. MODULE B — DealFlow AI: Finds matching global buyers in any region, qualifies each one for fit, and generates personalized cold outreach emails — ready to send in 20 seconds. MODULE C — Analytics Dashboard: Live risk heatmap, lead pipeline, and an AI-generated daily briefing with top 3 priorities for today. MODULE D — MarketPulse AI: Three-layer commodity price prediction — 24-48 hours (Tree-of-Thought reasoning), 1-5 year industry trends, and 5-10 year mega trend forecasts. Plus a Micro-Econ Agent that calculates the HHI monopoly index and applies Utility Maximization to optimize purchasing decisions within budget. AUTONOMOUS SYSTEM: A Vultr Cron job runs every night at midnight with zero human input — re-checking all tracked suppliers for new risks and cyber threats, updating the dashboard, and generating alerts while the user sleeps. Built with Featherless AI (DeepSeek-R1, Qwen 2.5 72B, Mistral Large — selected after benchmarking 600+ prompts), Kraken public market data API, and Vultr VM deployment on Amsterdam infrastructure. Live Demo: http://136.244.101.167:8000 GitHub: https://github.com/say18/tradenexus-ai-milan

CouncilOS

CouncilOS is a collaborative multi-agent reasoning platform that explores a more research-oriented alternative to traditional single-model AI workflows. Instead of generating a single isolated response, multiple AI models engage in human-like discussions where they critique, challenge, verify, and refine each other’s reasoning dynamically in real time. The system creates a live collaborative reasoning loop where models continuously inspect the same codebase, respond to each other’s observations, re-check execution logic, and progressively refine the analysis as the discussion evolves. To reduce repetitive or weak reasoning patterns, CouncilOS also applies semantic embedding-based filtering and execution-path validation to stabilize conversations dynamically. The current implementation focuses on code analysis, but the long-term vision extends toward a general-purpose collaborative reasoning infrastructure where multiple AI systems can work together to produce more reliable, multi-perspective, and transparent reasoning workflows.

MusKent Commerce OS

MusKent is a production-ready autonomous AI system designed to support real commerce operations across revenue, sales, automation, fulfillment, billing, and marketplace workflows. It moves beyond traditional copilots by combining reasoning agents, async execution, tool orchestration, and multimodal input into a unified operating system. At its core, MusKent uses agent-driven decision flows. It continuously evaluates business signals such as revenue performance, marketplace activity, sales trends, and operational state, then determines the next best action using AI. These agents operate within structured workflows, interact with internal tools and external APIs, and execute multi-step tasks while safely degrading to fallback systems when needed. The platform integrates multiple intelligence layers, including AI-powered reasoning for decision-making, specialized compute for ranking and scoring, and voice-based interaction through real-time and batch transcription. This enables a collaborative multi-agent system where different models and providers handle specific tasks like reasoning, analysis, and fallback execution. MusKent is designed for reliability and real-world usage. It supports asynchronous job processing for long-running operations, structured outputs for consistency, provider health awareness, and safe fallback mechanisms to maintain performance even under degraded conditions. From a systems perspective, MusKent delivers intelligent reasoning, agentic workflows, enterprise utility, and multimodal interaction in a single platform. The result is an AI-powered commerce operating system that can analyze, plan, and act across business operations while remaining resilient in production environments.

Shams AI

Shams is a local Al companion and autonomous observer designed to move beyond a chatbot into a real-world decision-making agent. Running on a Raspberry Pi, Shams combines voice interaction, a Fusion HAT microphone and speaker, an OLED animated face, camera capture, PIR motion detection, ultrasonic distance sensing, Telegram notifications, Gmail integration, and Ollama-powered local reasoning. It can detect events in the physical environment, respond with speech, take photos, send alerts, summarize unread email, remember tasks, and react to external automation events. Through UiPath integration, Shams can receive enterprise workflow events from a computer, such as long focus sessions, notification overload, idle time, or repetitive work patterns. Instead of only reporting data, Shams interprets the event and responds with a context-aware action or message. For example, if UiPath detects excessive multitasking, Shams can speak a calm intervention, log the event, notify the user, or trigger a follow-up workflow. The goal is to demonstrate an Al agent with presence: a system that listens, observes, remembers, reasons, and acts across bot the physical and digital world.

FuelGuard AI

FuelGuard AI is an AI-powered enterprise operational intelligence platform designed to help organizations monitor fuel operations, detect anomalies, and improve operational decision-making using Gemini AI. The platform combines real-time operational monitoring, anomaly detection, risk analysis, and AI-powered insights into a unified enterprise dashboard. FuelGuard AI uses Gemini-powered copilots to analyze operational patterns, explain risk behavior, generate executive summaries, and provide intelligent recommendations based on real operational data. Our system focuses on transforming operational data into actionable intelligence through AI-assisted analysis, anomaly investigation, and explainable risk assessment. The platform is designed for enterprise environments where visibility, operational awareness, and intelligent decision support are critical. FuelGuard AI integrates Angular, ASP.NET Core, PostgreSQL, and Gemini Flash to deliver a scalable cloud-ready architecture with AI-driven operational intelligence capabilities. The system includes AI operational chat, intelligent risk scoring, anomaly detection workflows, and enterprise analytics designed for real-world operational monitoring scenarios. The goal of FuelGuard AI is to move beyond traditional dashboards and create an enterprise AI copilot capable of helping organizations understand operational risks, identify suspicious behavior, and improve operational oversight through intelligent AI reasoning.