IBM Expands Granite Model Family with New Multi-Modal and Reasoning AI Built for the Enterprise
IBM has unveiled Granite 3.2, the next generation of its large language model family, focusing on small, efficient AI models for enterprise use. The new release includes:
- A vision language model (VLM) for document understanding that matches larger models' performance
- Enhanced reasoning capabilities in 2B and 8B models with switchable chain of thought functionality
- Slimmed-down Guardian safety models with 30% size reduction while maintaining performance
- New TinyTimeMixers (TTM) models for long-term forecasting up to two years
The models are available under Apache 2.0 license on Hugging Face and various platforms including watsonx.ai, Ollama, and Replicate. The Granite 3.2 8B model has shown significant improvements in instruction-following benchmarks and can rival larger models like Claude 3.5 Sonnet in math reasoning tasks.
IBM ha presentato Granite 3.2, la prossima generazione della sua famiglia di modelli linguistici di grandi dimensioni, focalizzandosi su modelli AI piccoli ed efficienti per l'uso aziendale. La nuova versione include:
- Un modello linguistico visivo (VLM) per la comprensione dei documenti che eguaglia le prestazioni dei modelli più grandi
- Capacità di ragionamento migliorate nei modelli 2B e 8B con funzionalità di catena di pensiero switchable
- Modelli di sicurezza Guardian snelliti con una riduzione del 30% delle dimensioni mantenendo le prestazioni
- Nuovi modelli TinyTimeMixers (TTM) per previsioni a lungo termine fino a due anni
I modelli sono disponibili sotto licenza Apache 2.0 su Hugging Face e varie piattaforme tra cui watsonx.ai, Ollama e Replicate. Il modello Granite 3.2 8B ha mostrato miglioramenti significativi nei benchmark di seguimento delle istruzioni e può competere con modelli più grandi come Claude 3.5 Sonnet in compiti di ragionamento matematico.
IBM ha presentado Granite 3.2, la próxima generación de su familia de modelos de lenguaje grande, centrada en modelos de IA pequeños y eficientes para uso empresarial. La nueva versión incluye:
- Un modelo de lenguaje visual (VLM) para la comprensión de documentos que iguala el rendimiento de modelos más grandes
- Capacidades de razonamiento mejoradas en modelos de 2B y 8B con funcionalidad de cadena de pensamiento intercambiable
- Modelos de seguridad Guardian reducidos con una reducción del 30% en tamaño manteniendo el rendimiento
- Nuevos modelos TinyTimeMixers (TTM) para pronósticos a largo plazo de hasta dos años
Los modelos están disponibles bajo la licencia Apache 2.0 en Hugging Face y varias plataformas, incluyendo watsonx.ai, Ollama y Replicate. El modelo Granite 3.2 8B ha mostrado mejoras significativas en los benchmarks de seguimiento de instrucciones y puede rivalizar con modelos más grandes como Claude 3.5 Sonnet en tareas de razonamiento matemático.
IBM은 Granite 3.2를 공개했습니다. 이는 대규모 언어 모델 가족의 차세대 버전으로, 기업 사용을 위한 작고 효율적인 AI 모델에 중점을 두고 있습니다. 새로운 릴리스에는 다음이 포함됩니다:
- 더 큰 모델의 성능과 일치하는 문서 이해를 위한 비전 언어 모델(VLM)
- 스위치 가능한 사고의 사슬 기능을 갖춘 2B 및 8B 모델에서의 향상된 추론 능력
- 성능을 유지하면서 30% 크기를 줄인 슬림화된 Guardian 안전 모델
- 최대 2년까지 장기 예측을 위한 새로운 TinyTimeMixers (TTM) 모델
모델은 Apache 2.0 라이선스 하에 Hugging Face 및 watsonx.ai, Ollama, Replicate를 포함한 다양한 플랫폼에서 제공됩니다. Granite 3.2 8B 모델은 지침 준수 벤치마크에서 상당한 개선을 보여주었으며, 수학적 추론 작업에서 Claude 3.5 Sonnet과 같은 더 큰 모델과 경쟁할 수 있습니다.
IBM a dévoilé Granite 3.2, la prochaine génération de sa famille de modèles de langage de grande taille, axée sur des modèles d'IA petits et efficaces pour un usage en entreprise. La nouvelle version comprend :
- Un modèle de langage visuel (VLM) pour la compréhension des documents qui égalise les performances des modèles plus grands
- Des capacités de raisonnement améliorées dans les modèles 2B et 8B avec une fonctionnalité de chaîne de pensée commutable
- Des modèles de sécurité Guardian allégés avec une réduction de taille de 30 % tout en maintenant les performances
- De nouveaux modèles TinyTimeMixers (TTM) pour des prévisions à long terme allant jusqu'à deux ans
Les modèles sont disponibles sous la licence Apache 2.0 sur Hugging Face et diverses plateformes, y compris watsonx.ai, Ollama et Replicate. Le modèle Granite 3.2 8B a montré des améliorations significatives dans les benchmarks de suivi des instructions et peut rivaliser avec des modèles plus grands comme Claude 3.5 Sonnet dans les tâches de raisonnement mathématique.
IBM hat Granite 3.2 vorgestellt, die nächste Generation seiner großen Sprachmodellfamilie, die sich auf kleine, effiziente KI-Modelle für den Unternehmenseinsatz konzentriert. Die neue Version umfasst:
- Ein visuelles Sprachmodell (VLM) zur Dokumentenverständnis, das die Leistung größerer Modelle erreicht
- Verbesserte Schlussfolgerungsfähigkeiten in den 2B- und 8B-Modellen mit umschaltbarer Denkweise-Funktionalität
- Verringerte Guardian-Sicherheitsmodelle mit einer Größenreduzierung von 30% bei gleichbleibender Leistung
- Neue TinyTimeMixers (TTM) Modelle für langfristige Prognosen von bis zu zwei Jahren
Die Modelle sind unter der Apache 2.0-Lizenz auf Hugging Face und verschiedenen Plattformen wie watsonx.ai, Ollama und Replicate verfügbar. Das Granite 3.2 8B-Modell hat erhebliche Verbesserungen bei den Anweisungsbefolgungsbenchmarks gezeigt und kann mit größeren Modellen wie Claude 3.5 Sonnet bei mathematischen Schlussfolgerungsaufgaben konkurrieren.
- New VLM matches performance of larger models while using fewer resources
- 8B model achieves double-digit improvements in instruction-following benchmarks
- 30% size reduction in Guardian safety models while maintaining performance
- Extended forecasting capabilities up to 2 years with sub-10M parameter models
- Apache 2.0 license enables broader enterprise adoption
- Chain of thought reasoning requires substantial compute power
- Performance improvements rely on additional compute-intensive inference scaling methods
Insights
IBM's release of Granite 3.2 represents a significant strategic advancement in the enterprise AI market, where the company is deliberately zigging while others zag. While competitors race to build ever-larger models, IBM is focusing on smaller, more efficient AI that delivers comparable performance at a fraction of the computational cost.
The technical achievements here are substantial. The new vision language model demonstrates that smaller models can match or exceed larger competitors (Llama 3.1 11B and Pixtral 12B) on document understanding tasks critical to enterprise workflows. This is particularly valuable for document-heavy industries like legal, financial services, and healthcare where processing efficiency directly impacts operational costs.
Most impressive is IBM's novel inference scaling approach that enables the 8B parameter model to rival much larger models like Claude 3.5 Sonnet or GPT-4o on complex math reasoning tasks. This represents a potential paradigm shift in the industry's approach to model scaling - focusing on inference optimization rather than simply increasing parameter count.
The ability to toggle reasoning capabilities on/off programmatically addresses a key enterprise concern: computational efficiency. This feature allows organizations to optimize for cost by using simpler processing for routine tasks while deploying more intensive reasoning only when needed, potentially reducing AI operational costs by 30-50% compared to always-on reasoning approaches.
IBM's Apache 2.0 licensing strategy contrasts sharply with more restrictive approaches from competitors, reducing legal friction for enterprise adoption and integration. This open approach, combined with wide availability across platforms, positions IBM to capture market share from enterprises concerned about vendor lock-in.
For investors, this release reinforces IBM's differentiated AI strategy focused on practical enterprise value rather than headline-grabbing capabilities. The company is building an ecosystem of specialized, efficient models that address specific business needs while maintaining favorable economics - a compelling value proposition as AI moves from experimentation to production at scale.
IBM's Granite 3.2 release represents a calculated strategic pivot in the enterprise AI race, focusing on operational efficiency rather than headline-grabbing model size. This approach directly challenges the prevailing "bigger is better" narrative while addressing the total cost of ownership (TCO) concerns that have hindered enterprise AI adoption.
The business implications are substantial. By delivering smaller models with comparable performance, IBM enables deployment on less expensive hardware with reduced energy consumption - potentially cutting infrastructure costs by 40-60% compared to larger models. This makes AI economically viable for a broader range of business processes and use cases that couldn't justify the expense of running massive models.
IBM's strategy creates clear differentiation against Microsoft's Copilot approach (which relies heavily on OpenAI's large models) and Google's Gemini offerings. While competitors are essentially renting access to their AI capabilities through API calls and cloud services, IBM's Apache 2.0 licensing and smaller models enable true on-premises deployment - a important advantage for industries with stringent data sovereignty requirements like healthcare, financial services, and government.
The toggle functionality for reasoning capabilities is particularly shrewd, allowing organizations to optimize for either speed or analytical depth depending on the task. This addresses a key pain point in enterprise AI adoption: the inability to right-size computational resources to business needs.
For IBM investors, this release reinforces the company's methodical approach to building an AI portfolio that complements its hybrid cloud strategy. Rather than competing directly with consumer-focused AI companies, IBM is carving out a defensible position in practical enterprise AI that integrates with existing business systems and processes.
The updated TinyTimeMixers models for long-range forecasting extend this practical approach to time-series applications - an underserved but highly valuable domain for business planning across supply chain, finance, and retail operations where even small improvements in forecasting accuracy translate directly to bottom-line impact.
- Granite 3.2 – small AI models offering reasoning, vision, and guardrail capabilities with a developer friendly license
- Updated Granite time series models that offer long-range forecasting with less than 10M parameters
All Granite 3.2 models are available under the permissive Apache 2.0 license on Hugging Face. Select models are available today on IBM watsonx.ai, Ollama, Replicate, and LM Studio, and expected soon in RHEL AI 1.5 – bringing advanced capabilities to businesses and the open-source community. Highlights include:
- A new vision language model (VLM) for document understanding tasks which demonstrates performance that matches or exceeds that of significantly larger models – Llama 3.2 11B and Pixtral 12B – on the essential enterprise benchmarks DocVQA, ChartQA, AI2D and OCRBench1. In addition to robust training data, IBM used its own open-source Docling toolkit to process 85 million PDFs and generated 26 million synthetic question-answer pairs to enhance the VLM's ability to handle complex document-heavy workflows.
- Chain of thought capabilities for enhanced reasoning in the 3.2 2B and 8B models, with the ability to switch reasoning on or off to help optimize efficiency. With this capability, the 8B model achieves double-digit improvements from its predecessor in instruction-following benchmarks like ArenaHard and Alpaca Eval without degradation of safety or performance elsewhere2. Furthermore, with the use of novel inference scaling methods, the Granite 3.2 8B model can be calibrated to rival the performance of much larger models like Claude 3.5 Sonnet or GPT-4o on math reasoning benchmarks such as AIME2024 and MATH500.3
- Slimmed-down size options for Granite Guardian safety models that maintain performance of previous Granite 3.1 Guardian models at
30% reduction in size. The 3.2 models also introduce a new feature called verbalized confidence, which offers more nuanced risk assessment that acknowledges ambiguity in safety monitoring.
IBM's strategy to deliver smaller, specialized AI models for enterprises continues to demonstrate efficacy in testing, with the Granite 3.1 8B model recently yielding high marks on accuracy in the Salesforce LLM Benchmark for CRM.
The Granite model family is supported by a robust ecosystem of partners, including leading software companies embedding the LLMs into their technologies.
"At CrushBank, we've seen first-hand how IBM's open, efficient AI models deliver real value for enterprise AI – offering the right balance of performance, cost-effectiveness, and scalability," said David Tan, CTO, CrushBank. "Granite 3.2 takes it further with new reasoning capabilities, and we're excited to explore them in building new agentic solutions."
Granite 3.2 is an important step in the evolution of IBM's portfolio and strategy to deliver small, practical AI for enterprises. While chain of thought approaches for reasoning are powerful, they require substantial compute power that is not necessary for every task. That is why IBM has introduced the ability to turn chain of thought on or off programmatically. For simpler tasks, the model can operate without reasoning to reduce unnecessary compute overhead. Additionally, other reasoning techniques like inference scaling have shown that the Granite 3.2 8B model can match or exceed the performance of much larger models on standard math reasoning benchmarks. Evolving methods like inference scaling remains a key area of focus for IBM's research teams.4
Alongside Granite 3.2 instruct, vision, and guardrail models, IBM is releasing the next generation of its TinyTimeMixers (TTM) models (sub 10M parameters), with capabilities for longer-term forecasting up to two years into the future. These make for powerful tools in long-term trend analysis, including finance and economics trends, supply chain demand forecasting and seasonal inventory planning in retail.
"The next era of AI is about efficiency, integration, and real-world impact – where enterprises can achieve powerful outcomes without excessive spend on compute," said Sriram Raghavan, VP, IBM AI Research. "IBM's latest Granite developments focus on open solutions demonstrate another step forward in making AI more accessible, cost-effective, and valuable for modern enterprises."
To learn more about Granite 3.2, read this technical article.
About IBM
IBM is a leading provider of global hybrid cloud and AI, and consulting expertise. We help clients in more than 175 countries capitalize on insights from their data, streamline business processes, reduce costs, and gain a competitive edge in their industries. Thousands of governments and corporate entities in critical infrastructure areas such as financial services, telecommunications and healthcare rely on IBM's hybrid cloud platform and Red Hat OpenShift to affect their digital transformations quickly, efficiently, and securely. IBM's breakthrough innovations in AI, quantum computing, industry-specific cloud solutions and consulting deliver open and flexible options to our clients. All of this is backed by IBM's long-standing commitment to trust, transparency, responsibility, inclusivity, and service. Visit www.ibm.com for more information.
Media contact:
Amy Angelini
IBM AI Communications
alangeli@us.ibm.com
1 | Vision model benchmark results are available in IBM's technical article, IBM Granite 3.2: Reasoning, Vision, Forecasting, and More, published February 26, 2025. |
2 | Instruct model benchmark results are available in IBM's technical article, IBM Granite 3.2: Reasoning, Vision, Forecasting, and More, published February 26, 2025. |
3 | Inference scaling benchmark results are available in IBM's technical research blog, Reasoning in Granite 3.2 Using Inference Scaling, published February 26, 2025. |
4 | Reasoning in Granite 3.2 Using Inference Scaling, IBM, published February 26, 2025. |
View original content to download multimedia:https://www.prnewswire.com/news-releases/ibm-expands-granite-model-family-with-new-multi-modal-and-reasoning-ai-built-for-the-enterprise-302386080.html
SOURCE IBM
FAQ
What are the key features of IBM's new Granite 3.2 AI model?
How does IBM Granite 3.2's performance compare to larger AI models?
Where can developers access IBM's Granite 3.2 models?
What improvements does IBM Granite 3.2 offer for document processing?
How does IBM's Granite 3.2 optimize computational efficiency?
