Arm has launched CSS for Mobile 2, a heterogeneous compute platform that pairs a new C2-Ultra CPU with Scalable Matrix Extension 2 (SME2) and the Mali G2-Ultra NX GPU. Arm cites up to 1.7x performance gains across recent AI models and a 24% reduction in completion time for representative end-to-end agentic workflows versus the previous generation.

The platform targets agentic AI workloads that translate intent into action across multiple compute engines within a single mobile device. Arm positions the CPU, GPU, system IP, physical implementation, and software as a single configurable bundle, while preserving the option to license individual components for custom SoC designs.

What does the C2 CPU cluster deliver?

The C2 CPU cluster combines C2-Ultra and C2-Pro cores with two SME2 units, doubling SME2 capability compared with the prior generation. Performance figures cited by Arm include:

• Up to 1.7x performance improvement across the latest AI models
• Up to 15% higher single-thread performance
• 15% faster web browsing
• 12% faster application launch
• 12% higher multi-thread performance at the cluster level
• Up to 70% peak uplift in selected AI tasks

Arm measures a 24% faster completion time on a representative agentic workflow spanning speech processing, memory retrieval, reasoning, application execution, and web browsing. Higher single-thread performance cuts latency for individual decisions, the company says, while higher multi-thread performance keeps search, models, and applications progressing in parallel.

The cluster also serves as the orchestration layer. It maintains task state, assembles context, and decides where each stage should run — local applications, on-device models, or cloud services — while coordinating permissions, decisions, and the progression of the task.

How does the Mali G2-Ultra NX change mobile graphics?

The Mali G2-Ultra NX GPU integrates dedicated neural accelerators directly into the shader cores, adds a new execution engine, and introduces third-generation ray tracing. Arm describes this combination as the foundation for neural graphics workloads running natively on the GPU, complementing the C2 cluster's agentic AI capability.

vivo has signed on as a launch partner. The smartphone maker said: "Together with Arm, we are bringing Arm Neural Technology to the vivo's latest flagship smartphones, built on Arm's latest compute platform, to unlock new possibilities for mobile gaming. By combining Arm's latest compute and graphics technologies with vivo's device-level optimizations, we aim to accelerate neural graphics adoption and deliver more immersive gaming experiences on mobile."

How does the platform handle data movement?

The SI L2 system interconnect provides lower-latency access, higher platform bandwidth, coherency, and quality-of-service features. These capabilities manage and prioritize activity across CPU, GPU, and other resources as they operate concurrently and share access to memory.

What tools support developers?

Arm ties the new hardware to a software ecosystem used by more than 22 million developers. KleidiAI and framework integrations expose SME2 acceleration through familiar AI stacks, so developers gain the underlying CPU capability through software they already use.

Arm AI Portal adds a central location for validated, optimized models, along with performance and accuracy data, code samples, and deployment resources. The Arm MCP Server then brings those capabilities into agentic development environments, shortening the path from model selection to on-device deployment for workloads targeting CSS for Mobile 2.

What flexibility do silicon partners retain?

Partners can license individual components, combine them with custom or third-party IP, or adopt the full configuration. CSS for Mobile 2 also extends optimization into physical implementation, allowing partners to weigh performance, power, and area before starting SoC integration. That system-level scope defines how Arm expects differentiation to happen across the next mobile cycle.