Arm Create: AI Optimization Challenge

Overview

The Arm Create: AI Optimization Challenge 2026 encourages developers to build and submit projects that showcase how AI can be optimized for Arm-powered platforms. Participants can choose from three distinct tracks: Physical AI, Cloud AI, and Mobile AI, each focusing on different optimization frontiers. The Physical AI track targets real-world systems like robotics, embedded devices, and autonomous systems, requiring AI inference to run on-device or nearby edge compute. The Cloud AI track focuses on scalable infrastructure, including Arm64 cloud, inference performance, and production-ready developer workflows. Meanwhile, the Mobile AI track emphasizes optimizing AI for on-device constraints such as performance, privacy, latency, and battery efficiency on Arm-powered client devices. Across all tracks, submissions must demonstrate clear optimization work and, where possible, measurable improvements. This includes enhancing model size, quality, speed, inference server speed, developer experience, and implementing Arm-specific optimizations. Developers are encouraged to utilize Arm Performix for benchmarking their Arm-based performance to clearly present their results. The challenge aims to foster innovation in AI optimization on Arm architecture, providing a platform for developers to explore new possibilities and contribute to the Arm ecosystem. Winning projects will not only receive cash prizes but also be featured in the Arm Community Blog.

What to build

Entrants must create, migrate, or optimize an AI solution on Arm architecture, falling within at least one of three specified tracks: Physical AI, Cloud AI, or Mobile AI. Submissions should demonstrate clear optimization work and measurable improvements. Optimizations can focus on model size reduction, improved model quality, increased model speed (e.g., tokens/sec), enhanced inference server speed, better developer experience (tools, workflows, documentation), or Arm-specific optimizations to make existing frameworks, libraries, models, or applications run better on Arm. Developers can use Arm Performix to benchmark performance.

Prizes

Key dates

• Registration Period BeginsJun 10, 2026 9:00 am Pacific Time
• Submission Period BeginsJun 10, 2026 9:00 am Pacific Time
• Office HoursJul 14, 2026 10:00 AM PDT
• Office HoursAug 3, 2026 10:00 AM PDT
• Registration Period EndsAug 14, 2026 4:00 pm Pacific Time
• Submission Period EndsAug 14, 2026 4:00 pm Pacific Time
• Judging Period BeginsAug 17, 2026 9:00 am Pacific Time
• Judging Period EndsSep 4, 2026 4:00 pm Pacific Time
• Winners AnnouncedSep 15, 2026 2:00 pm Pacific Time

Eligibility

The Hackathon is open to individuals who are at least the age of majority in their country of residence, teams of eligible individuals, and organizations (corporations, non-profits, LLCs, partnerships) that exist and are organized/incorporated at the time of entry. Participants from all countries/territories are eligible, excluding those where participation or prize receipt is prohibited by US or local law (e.g., Brazil, Quebec, Russia, Crimea, Cuba, Iran, North Korea, and other OFAC-designated countries). Employees, representatives, and agents of promotion entities, judges, and their immediate families or households, or any other individual/organization creating a conflict of interest are not eligible.

How to enter

1. Register for the Hackathon on the Hackathon Website by clicking the “Join Hackathon” button. Sign up for a free Devpost account or log in with an existing one.
2. Create a free developer account with Arm AI by signing up for the Arm Developer Program to access a personalized dashboard, on-demand labs, technical support, and developer tools.
3. Complete the Arm Developer Program Sign-Up Form, including Name, Company, Email Address, and Job Title. Verify your email with a code.
4. Complete your Arm Developer Program profile (or update an existing Arm ID) and log in to your Arm Developer Dashboard.
5. Create, migrate, or optimize an AI solution on Arm architecture that falls within one of the three challenge tracks (Physical AI, Cloud AI, or Mobile AI).
6. Complete and enter all required fields on the “Enter a Submission” page of the Hackathon Website during the Submission Period, adhering to all Project and Submission Requirements.

Submission requirements

• A project built with required developer tools and meeting Project Requirements.
• A public and open-source URL to your code repository (e.g., GitHub) containing all necessary source code, assets, and instructions for the project to be functional. The repository must include an MIT or Apache 2.0 open-source license file, visible at the top of the page.
• A text description explaining the features and functionality of your Project, including: Project Overview (brief description, purpose, what makes it interesting), Functionality / Output (what the project does, final output), and Setup Instructions (step-by-step instructions to build/run/validate on an Arm-powered device or Arm64 environment).
• For Track 1 & Track 2: A copy of the project’s source code, either attached directly or linked to an open-source repository (e.g., GitHub).
• For Track 3: Proof artifacts (links/screenshots) as described in the track requirements.
• Optional: A demonstration video (under three minutes, showing the Project functioning on the target device, uploaded to YouTube, Vimeo, or Youku, and publicly visible). Must not include third-party trademarks or copyrighted material without permission.
• The Project must be capable of successful installation and consistent running on its intended platform, functioning as depicted/described.
• If the Project existed prior to the Hackathon, it must have been significantly updated after the start of the Submission Period, with an explanation of updates.
• If integrating third-party SDKs, APIs, or data, entrants must be authorized to use them according to terms and conditions/licensing.
• Access must be provided to a working Project for judging and testing via a link to a website, functioning demo, or test build. If private, login credentials must be included.
• All submission materials must be in English, or provided with an English translation.
• Each submission must clearly select one of the three tracks (Track 1, Track 2, or Track 3).

Judging criteria

• Technological Implementation – 40 points: Quality of software development, clear leverage of Arm-powered platforms (on-device, Arm64, efficiency-minded design), sound and well-executed technical approach.
• User Experience / Developer Experience – 15 points: Clarity of use/run/validation, structured documentation, potential for further development or reuse by other developers.
• Potential Impact – 20 points: Usefulness to the developer community, creation of reusable artifacts (optimized models, migration templates, prompt assets, learning-ready content).
• “WOW” factor – 25 points: Creativity and compelling nature of the submission, outstanding approach, usefulness, or clarity, ability to quickly capture attention and communicate value.

Sponsors & partners

Resources

• Arm Developer Program
• Arm Learning paths
• Arm Developer Ecosystem GitHub
• Arm Developer Program Discord
• Arm Performix
• Full Rules
• Physical AI - Arm Market Page
• Cloud AI - Arm Market Page
• Mobile AI - Arm Market Page
• Track Details Tab
• Arm Developer Program Sign-Up Page
• ROS 2 install guide (Physical AI learning path)
• Cyclone DDS install guide (Physical AI learning path)
• Robot Simulation and Reinforcement Learning Workflows on Arm (Physical AI learning path)
• Deploy multi-node Zenoh systems on Raspberry Pi (Physical AI learning path)
• Deploy Open AD Kit containerized autonomous driving simulation on Arm Neoverse (Physical AI learning path)
• Prototype safety-critical isolation for autonomous driving systems on Arm Neoverse (Physical AI learning path)
• Visualize Ethos-U NPU performance with ExecuTorch (Physical AI learning path)
• Deploy an AI Agent on Arm with llama.cpp and llama-cpp-agent using KleidiAI (Cloud AI learning path)
• Deploy a Large Language Model (LLM) chatbot with llama.cpp using KleidiAI on Arm servers (Cloud AI learning path)
• Deploy a LLM-based Vision Chatbot with PyTorch and Hugging Face Transformers on Google Axion processors (Cloud AI learning path)
• Accelerate Whisper on Arm with Hugging Face Transformers (Cloud AI learning path)
• Run vLLM inference with INT4 quantization on Arm servers (Cloud AI learning path)
• Run a Large Language Model (LLM) chatbot with PyTorch using KleidiAI on Arm servers (Cloud AI learning path)
• Run an LLM chatbot with rtp-llm on Arm-based servers (Cloud AI learning path)
• Deploy SqueezeNet 1.0 INT8 model with ONNX Runtime on Azure Cobalt 100 (Cloud AI learning path)
• Deploy TensorFlow on Google Cloud C4A (Arm-based Axion VMs) (Cloud AI learning path)
• Build a Multimodal Retail Restocking Assistant on Armv9 With MNN (Cloud AI learning path)
• Build an Android chat app with Llama, KleidiAI, ExecuTorch, and XNNPACK (Mobile AI learning path)
• Build a customer support chatbot on Android with Llama and ExecuTorch (Mobile AI learning path)
• Run LLM inference on Android with KleidiAI, MediaPipe, and XNNPACK (Mobile AI learning path)
• Measure LLM inference performance with KleidiAI and SME2 on Android (Mobile AI learning path)