IBM Releases Granite 4.2 Models With Native Reasoning
IBM released Granite 4.2 on August 25, adding 3B, 8B and 30B open language models that it says are built for enterprise agents. The dense models add reasoning, tool use and coding capabilities under an Apache 2.0 license, giving teams another open-weight option to test for multi-step workflows.
IBM has released Granite 4.2, a family of open language models it says is designed for enterprise agents that need to reason through multi-step work before acting. The release includes 3B, 8B and 30B language models, and IBM positions the family around reasoning, tool use, coding, instruction following and speech capabilities.
What IBM released
IBM says the new language models use a multi-stage reinforcement-learning approach. Its stated goal is to help the models plan, weigh choices and verify a course of action in agentic workflows rather than simply respond to a prompt. The models are available under Apache 2.0, which allows teams to download, adapt and deploy them without proprietary licensing restrictions.
The company is also emphasizing dense architectures. That matters for deployment because dense models can be used across a wider range of infrastructure, while the available sizes let teams match a model to a high-throughput task or a more demanding reasoning and coding workload.
What remains to be tested
The release describes intended capabilities and IBM's training approach; it is not an independent benchmark of production reliability. Teams considering Granite 4.2 should therefore test the model on their own tool-calling, latency, safety and task-completion requirements before treating the announcement as a performance guarantee.
The practical change is that IBM is offering an open-weight option explicitly framed around enterprise agents, including workflows that require tools and multi-step planning. That gives organizations another model family to evaluate where deployment flexibility and permissive licensing are material constraints.
Key Points
- 1IBM released Granite 4.2 in 3B, 8B and 30B language-model sizes on August 25.
- 2IBM describes the models as dense, open-weight models with reasoning, tool use, coding and instruction-following capabilities.
- 3The release is under Apache 2.0; production reliability and task fit still require independent evaluation.
Scoring Rationale
A same-day open-model release from IBM with material relevance to teams evaluating enterprise agents, tool use and self-hosted or adaptable language models. The story is limited to IBM's stated release details and does not treat vendor claims as independent performance evidence.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
