Arm Launches AI Portal to Unify Developer Access to Optimized Software on Arm Platforms

At the Arm Create developer conference in Shanghai on September 11, 2026, Arm officially launched the Arm AI Portal, a unified entry point for developers and agents to discover, optimize, and deploy AI software on Arm compute platforms. The portal is now live.

The portal's first batch of pre-optimized models includes Alibaba's Qwen, Google Gemma, and Ultralytics YOLO, with support for runtimes such as ExecuTorch, LiteRT, and ONNX-RT. Ecosystem partners include Alibaba, Raspberry Pi, and Ultralytics. Developers can search task-specific models, access performance and accuracy data, compare latency, memory footprint, and model size, and use code examples and deployment workflows.

Looking ahead, Arm AI Portal will add tools that allow developers to import and optimize their own models, including proprietary ones, on Arm platforms. Early access resources for agentic AI are available now, with a full release to follow. Optimized Arm models can be accessed via Hugging Face, and Portal resources are available to coding agents through the Model Context Protocol (MCP).

In performance tests cited by Arm, Qwen3-TTS running with single-thread execution and a mixed-quantization configuration using a Q8_0 talker and code predictor achieved a more than 4x speedup on the vivo X300 with Arm's second-generation Scalable Matrix Extension (SME2). Ultralytics YOLO26 delivered over 40% performance gains on the vivo X300 with SME2 and FP16 vs FP32, and similarly on Raspberry Pi 5 with NEON and mixed FP16/INT8 quantization. Arm AI Portal covers Arm compute platforms across cloud, edge, and physical AI, helping Arm's ecosystem of more than 22 million developers and agents discover and use software optimized for their target hardware.