IBM has released the updated Granite 4.2 language models in 3B, 8B, and 30B parameter sizes, purpose-built for enterprise agentic workflows with native step-by-step reasoning capabilities. Alongside the language models, IBM introduced Granite Speech 5.0 Turbo CTC, a lightweight speech model optimized for high-throughput transcription on edge devices.

  • Granite 4.2 features dense architecture and multi-stage reinforcement learning, including "agentic RL" for software engineering and terminal-based coding tasks.
  • Models were trained on 1 trillion tokens of synthetic code from IBM’s CodeAlchemy pipeline and include speculative decoding to improve inference speed and reduce costs.
  • Granite Speech 5.0 Turbo CTC has 470 million parameters, uses connectionist temporal classification (CTC) for efficient streaming audio, and achieves an RTFx score of approximately 12,600 on H100 GPUs.
  • Both language and speech models are available under the Apache 2.0 license, allowing organizations to download, fine-tune, and deploy them without licensing restrictions.

These updates aim to help enterprises build reliable agents that can plan, execute complex tasks, and process voice data efficiently in real-world scenarios.