IBM has released Granite 4.2, a new family of reasoning AI models.
They come in three sizes and support thinking modes.
The larger models can use tools and act as agents.
What happened
IBM released Granite 4.2, its first family of reasoning large language models (LLMs, AI systems that understand and generate text), in three sizes: 3B, 8B, and 30B. The models are pre-trained from scratch on roughly 15T tokens and extend the context window to 512K tokens.
Why it matters
The 8B and 30B models add agentic training, meaning they can operate as agents that use tools, edit and run code, and search the web. All models include a thinking/non-thinking switch and native tool calling, making them practical for business automation and complex tasks.
What to watch
The models are released under the Apache 2.0 license, and the larger models (8B and 30B) can act as agents inside real sandboxed environments, which could enable more autonomous AI workflows.
Ask the AI about this article →
Earlier Granite releases were strong instruction-following assistants, but Granite 4.2 adds explicit reasoning capabilities. The training pipeline is extensive: pre-training on 15T tokens, supervised fine-tuning with chain-of-thought data, and a multi-stage reinforcement learning process. The 8B and 30B models receive an additional agentic RL block that enables them to use tools in real sandboxed environments, a significant step beyond typical LLM releases. This means businesses can deploy models that not only reason but also act, taking actions like writing code or searching the web, all under a permissive license.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Meta Platforms will launch its AI agent Hatch in the coming weeks and release a new AI model called Watermelon…
Snowflake says its Observe platform's AI SRE helps customers troubleshoot up to 10x faster, with an average of…
OpenAI has integrated ChatGPT with Apple Messages, allowing it to search messages, catch you up on conversatio…
Apple has refreshed its Mac mini and Mac Studio desktop computers, specifically designed for local AI developm…
OpenAI announced its new AI chip, Jalapeño, which delivers faster responses and higher throughput compared to…
At the Hot Chips conference on Tuesday, OpenAI shared the first benchmark results for its Jalapeño chip
