Qwen3-Coder

Top Builders

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

Qwen3-Coder

Qwen3-Coder is Alibaba Cloud's dedicated coding model, released on July 22, 2025. The flagship variant, Qwen3-Coder-480B-A35B-Instruct, is a mixture-of-experts model trained on 7.5 trillion tokens across 358 programming languages, with 70% of the training data being code. It supports 256K tokens of native context, extensible to 1M tokens with extrapolation, making it suited to repository-scale tasks and multi-file agentic workflows.

General
Release date	22 Jul 2025
Developer	Qwen / Alibaba Cloud
Type	Open-weight coding LLM (MoE)
License	Apache 2.0
GitHub	QwenLM/Qwen3-Coder
Hugging Face	Qwen3-Coder-480B-A35B-Instruct
Documentation	qwenlm.github.io/blog/qwen3-coder

Core Features

480B/35B MoE architecture: 480B total parameters with 35B active per token, using 160 experts with 8 activated per inference step.
256K native context: natively processes 256,000 tokens, with extrapolation support up to 1,000,000 tokens.
358 programming languages: trained on a broad code corpus covering mainstream and niche languages.
Agent RL post-training: long-horizon reinforcement learning trains the model to solve real-world tasks through multi-turn tool interactions.
Apache 2.0: weights are available for commercial use and fine-tuning.

Benchmarks

Benchmark	Score
SWE-Bench Verified	69.6%

SWE-Bench Verified scores are state-of-the-art among open models at release, comparable to Claude Sonnet 4.

Tools and Resources

Qwen3-Coder Blog Post: release announcement with benchmark details.
GitHub (QwenLM/Qwen3-Coder): model code and usage examples.
Hugging Face: model weights.
Qwen Code: open-source terminal agent built on Qwen3-Coder.
Together AI: hosted API access.

Ecosystem and Integrations

Paired with Qwen Code, an open-source terminal coding agent with GitHub Actions support.
Accessible via the Alibaba Cloud DashScope API using an OpenAI-compatible endpoint.
Available on Together AI, LM Studio, and Ollama for local and cloud inference.

To use Qwen3-Coder via API, get an API key on the Qwen API Platform. For local agentic coding, see the Qwen Code terminal agent.

Edit on GitHub

Qwen Qwen3-Coder AI technology Hackathon projects

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

Apohara VOUCH

Apohara VOUCH turns multi-agent decisions into cryptographically-verifiable offline receipts — signed, hash-chained, timestamped, and audit-ready in under 30 seconds. Built on 3 production LLM sponsors (Band SDK + AI/ML API + Featherless AI) with a deterministic post-LLM gate (BAAAR) that fails-closed on five auditable halt conditions. EU AI Act Art. 12 by construction. **When AI agents make a regulated decision, you can't trust the decision — and you can't prove it either.** Procurement, lending, hiring, and customer escalation are now mediated by multi-agent systems: 5–10 LLMs coordinate through chat rooms, hand off state, vote, and reach a verdict. Three failures follow: 1. **No audit trail.** When a regulator asks "who decided this, and why?", you have a chat log — not an evidence packet. Logs can be edited. Screenshots can be forged. LLM weights are opaque. 2. **No failure mode.** The agents coordinate, but if one hallucinates a vendor ID, the room reaches the wrong verdict anyway. Multi-agent consensus is consensus on the wrong answer. 3. **No offline verifiability.** The regulator asks for proof. You re-run the agents. They produce a different answer. The room is no longer reproducible. The EU AI Act Art. 12 (record-keeping), DORA Art. 16 (ICT incident logs), NIST AI RMF (Manage), and OWASP Agentic all require verifiable, tamper-evident, offline-checkable evidence. None of the existing solutions — vector stores, prompt logs, evals — satisfy all three. **Apohara VOUCH** is the first multi-agent substrate that produces EU AI Act Art. 12 evidence packets by construction, verified offline in under 30 seconds, with no LLM in the critical path. **Apohara VOUCH — vouch for every agent decision.**

Quorum: Governed Business Intelligence

Quorum is a Band-powered governed analytics platform that turns natural-language business questions into trustworthy, auditable answers. Instead of relying on a single AI model, Quorum uses a council of specialized agents that collaborate through a structured governance workflow. A Planner creates an investigation strategy, a Plan Guardian reviews the approach before execution, a SQL Analyst generates and executes queries, a Cost Sentinel validates cost and plan compliance, a Governance Guardian reviews correctness after execution, and a Decision Reporter converts findings into actionable recommendations. Quorum supports two investigation modes. Governed Chain answers descriptive questions such as "What happened?" through a sequential review process. Investigation Board tackles diagnostic questions such as "Why did revenue decline?" by launching multiple investigators in parallel and using an adjudicator to evaluate competing explanations before reaching a conclusion. Built on Band's multi-agent framework, Quorum enables structured collaboration between agents, allowing them to plan, review, challenge, validate, and justify outputs before a result reaches the user. This creates a transparent decision-making process rather than a black-box AI response. A core innovation is Quorum's governance-first design. Every plan, SQL query, agent decision, cost estimate, compliance review, revision request, and final recommendation is captured in a complete audit trail. Users can inspect investigation timelines, generated SQL, governance decisions, and supporting evidence behind every answer. The platform combines FastAPI, Next.js, PostgreSQL/SQLite, and LiteLLM-based model routing with support for Groq, AI/ML API, Featherless, OpenAI-compatible providers, and Ollama. By combining Band's collaborative agent orchestration with enterprise governance controls, Quorum delivers explainable AI-powered analytics that organizations can confidently use for business decision-making.

Bandwith

Welcome to Bandwidth (originally conceptualized for the Band of Agents Hackathon). Bandwidth is a multi-agent AI orchestration framework designed to revolutionize the software development lifecycle. By treating specialized AI models like members of a synchronized musical band, Bandwidth delegates complex engineering tasks to a unified digital development team. What is Bandwidth? Modern software development requires juggling architecture, coding, debugging, and testing. Bandwidth acts as the "conductor," managing a suite of specialized AI coding agents that work in parallel. Instead of relying on a single AI assistant to do everything sequentially, you deploy a full "band" where each agent is an expert in its specific domain—whether that's writing front-end components, optimizing database queries, or generating robust unit tests. Key Features - Multi-Agent Orchestration: Seamlessly coordinate multiple AI agents working on different parts of your codebase simultaneously. - Specialized Agent Roles: Assign specific tasks to dedicated agents (e.g., Lead Developer, QA Tester, DevOps Engineer) to ensure high-quality, focused output. - Automated Synchronization: The central conductor agent ensures that all generated code is harmonized, tested, and ready for deployment without painful conflicts. - Massive Throughput: Dramatically increase your team's development capacity—your "bandwidth"—by offloading boilerplate, testing, and routine feature development to the agent ecosystem. Whether you're a solo developer looking to multiply your output or a startup aiming to eliminate development bottlenecks, Bandwidth provides the framework to build faster, smarter, and perfectly in sync.

Ferry: Agentic Legacy Modernization

Ferry is an AI-assisted legacy modernization platform that helps teams move old systems into modern codebases with more visibility and control. Instead of handing an entire migration to one model, Ferry coordinates a Band room of specialized agents that each own a clear part of the process: planning, source analysis, business logic extraction, target code generation, database migration, test generation, review, final approval, and GitHub pull request creation. A user points Ferry at a legacy COBOL, Java, or PHP repository. Ferry analyzes the project, translates the application to Go or Rust, prepares MySQL MyISAM to InnoDB migration plans when needed, writes tests against the original behavior, and produces migration artifacts the team can review. The workflow ends in a GitHub pull request, so teams can inspect the diff, tests, and migration plan before anything is merged.

GridAI - DER Coordination Protocol

Problem Australia has roughly 15 GWh of home batteries and the number is climbing fast. They mostly see the same thing: the National Electricity Market price signal. When price drops in the evening they all discharge at once. Following one shared signal makes a fleet synchronise, and a synchronised fleet builds a new evening demand peak instead of smoothing the old one, while pushing voltage outside legal limits at the edge of the distribution network. This failure mode gets worse as virtual-power-plant deployment scales, because it appears precisely when fleets start coordinating against shared signals. It is a second-order problem that today's market design walks straight into. GridAI's novelty is the diagnosis: desynchronisation depends on fleet-level value heterogeneity, and each voltage breach can be attributed by cause, separating PV-export conditions from battery-herding events so only protocol-induced failures escalate. Solution GridAI is a multi-agent coordination protocol. Four agents, Forecaster, Coordinator, Compliance, and Operator, collaborate through Band as the actual collaboration layer, not a notification wrapper. The Coordinator runs a priority-based dispatch: each battery's slot is allocated from global fleet state using its state of charge and the owner's willingness-to-discharge. The fleet desynchronises through heterogeneity, the diversity in what each battery wants, not through symmetric negotiation. The Compliance agent reviews every plan against AS IEC 60038:2022 voltage limits, flags battery-herding breaches (kept distinct from midday PV-export breaches), and escalates to a human Operator with a full Band-native audit trail. Result: battery-herding overvoltage breaches cut from 471 to 0, fleet synchrony from 1.000 to 0.167. Convergence takes 1 to 2 rounds, runs on existing inverter hardware, and fits the CSIP-AUS standard already mandated in Australia.

AegisMesh: Multi-Agent Security Remediation

AegisMesh is a Band-powered multi-agent security platform that transforms vulnerability remediation from a manual process into an autonomous, explainable workflow. Traditional security scanners identify vulnerabilities but leave developers responsible for designing fixes, validating those fixes, and determining whether the application is actually secure. AegisMesh addresses this challenge using three specialized AI agents coordinated through Band. The Blue Coder Agent generates remediation patches for vulnerable code. The Red Auditor Agent performs adversarial security analysis using Graph-of-Thought reasoning to challenge those patches and search for exploit paths. The Security Intelligence Agent evaluates the overall security posture and produces a final risk assessment and recommendation. Band serves as the coordination layer between agents, enabling role specialization, task handoffs, shared context, task state management, transcript generation, and workflow orchestration. Vulnerabilities move through a structured lifecycle including triage, patch generation, adversarial validation, and security intelligence reporting. AegisMesh emphasizes explainability and auditability. Users can inspect attack paths evaluated by the auditor, review reasoning behind findings, and explore complete agent transcripts that show how remediation decisions were reached. The platform also includes an AI/ML intelligence dashboard that provides visibility into model usage, token consumption, request volume, and inference costs during execution. The system combines frontier AI models selected for their strengths: Qwen3-Coder for patch generation, DeepSeek for adversarial auditing, and GPT-4o for security intelligence. By combining specialized AI agents coordinated through Band, AegisMesh moves beyond vulnerability detection to deliver autonomous remediation, adversarial validation, explainable reasoning, and actionable security intelligence in a single workflow.