IBM has launched Granite 4.2, the latest version of its open-weight large language models designed for self-hosting. The models come in 3B, 8B, and 30B parameter variants with a 128,000-token context window. The release focuses on reasoning capabilities through chain-of-thought processing, allowing more rigorous responses but requiring higher compute resources. The 8B and 30B versions include agentic reinforcement learning for tasks like terminal use and web searching. IBM positions Granite as prioritising predictable enterprise deployments rather than cutting-edge performance. The models appeal to organisations seeking local alternatives to cloud-based services from Anthropic or OpenAI, offering cost savings and customisation options. The release addresses growing interest in local models amongst enterprises, developers, and hobbyists seeking to avoid per-token API fees whilst maintaining control over AI deployments.

IBM and the United States Tennis Association have unveiled new AI-powered features for the 2025 US Open app and website. The innovations include a Live Updates homepage for personalised content, and Serve Quality, a first-to-market metric using advanced limb-tracking technology to analyse serve mechanics across all 254 singles matches. The tool examines 21 joint points on players' bodies, capturing 50 snapshots per second and analysing 1.2 billion data points throughout the tournament. Other enhancements include Key Moments, which explains momentum shifts in matches, and an improved Match Chat feature powered by watsonx Orchestrate. IBM research found 72% of surveyed fans use sports apps as their central hub for following events, whilst 46% report increased expectations for digital sports experiences.

IBM announced a dual-architecture processor for future IBM Z and LinuxONE systems that will run both IBM and Arm applications. The chip, unveiled at the Hot Chips conference, enables enterprises to operate Arm-native Linux environments alongside z/OS and Linux on IBM Z. Built on 2 nanometre technology, the processor features 11 high-performance cores running above 5.7 GHz, AI inference accelerators for fraud detection, and an on-chip data processing unit. Each core can natively execute both Arm and IBM instructions concurrently. The development stems from IBM and Arm's April 2026 collaboration. It brings Arm's 22 million-strong developer ecosystem to IBM's mainframe platforms whilst maintaining IBM's security, reliability, and encryption capabilities. IBM Z and LinuxONE platforms can scale to hundreds of cores and tens of terabytes of memory.