Welcome to This Week’s AI Insights!
As AI evolves rapidly, staying ahead of the latest developments is crucial. In this newsletter edition, we bring you groundbreaking updates from IBM’s Granite 3.2, OpenAI, and Qualcomm, shaping the future of AI innovation and enterprise adoption.
IBM has unveiled Granite 3.2, a smaller, more powerful AI model that enhances reasoning, multimodal capabilities, and cost efficiency.
Additionally, the launch of AI agents powered by watsonx makes AI-driven automation more accessible. However, with the rise of autonomous AI agents, governance has become essential.
IBM has introduced new governance and evaluation tools in Watsonx. Governance to ensure responsible AI usage. Qualcomm and IBM have expanded their collaboration to scale AI from edge to cloud, improving enterprise AI solutions in a significant industry move.
Meanwhile, OpenAI is increasing access to GPT-4.5, its latest and most advanced model, though questions remain about its long-term API availability due to high operational costs.
IBM Launches Smaller AI Model With Advanced Reasoning
Newly introduced IBM’s Granite 3.2, the latest version of its AI model family. The update brings enhanced reasoning, multi-modal AI, and improved cost efficiency.
IBM’s Granite 3.2 includes a vision language model (VLM) that processes documents, classifies data, and extracts information. IBM claims the VLM performs as well as, or better than, larger models like Llama 3.2 11B and Pixtral 12B.
The model also features new reasoning techniques, such as inference scaling. These allow the Granite 3.2 8B model to match or exceed larger models on standard math reasoning benchmarks.
Some versions of IBM’s Granite 3.2 include chain-of-thought reasoning, where AI outlines its thinking process. This feature can be switched on or off to save costs.
IBM is positioning Granite 3.2 as a more efficient and accessible business AI solution. The model was trained using IBM’s open-source Docling toolkit, which helps convert documents into specialized data for enterprise AI.
It processed 85 million PDFs and 26 million synthetic question-answer pairs to enhance document handling capabilities.
Granite 3.2 also includes smaller versions of the Granite Guardian safety models. These maintain the performance of Granite 3.1 while reducing size by 30%.
A new “verbalized confidence” feature improves risk assessment by acknowledging uncertainty.
IBM is making the models available under the Apache 2.0 license on Hugging Face. Select versions are also available on IBM watsonx.ai, Ollama, Replicate, and LM Studio, with RHEL AI 1.5 support coming soon.
Alongside Granite 3.2, IBM is launching the next generation of its TinyTimeMixers (TTM) models.
These compact AI models are designed for long-term time-series forecasting, with predictions extending up to two years.
With these updates, IBM continues to push for efficient and cost-effective AI adoption in enterprises.
If you want to dig further, click here.
AI Agents Built with watsonx
IBM has introduced two new integrations within watsonx: Lamatic.ai and Serenity Star. These platforms make building AI agents powered by Granite models hosted on watsonx easier.
What is Lamatic.ai?
Lamatic.ai is a fully managed platform-as-a-service (PaaS) with a low-code visual builder. It features integrated vector stores and seamless connections to apps, data sources, and AI models, allowing users to rapidly build, test, and deploy high-performance AI agents at the edge.
What is Serenity Star?
Serenity Star provides AI-driven solutions for automated content creation, product development, and customer service. It specializes in industry-specific algorithms designed to boost productivity and competitiveness.
Its AI agent, Serena, can work autonomously to:
- Collect data
- Execute predefined tasks
- Achieve user-defined objectives
With these integrations, watsonx expands its capabilities, making AI agent development more accessible and efficient for businesses.
Visit the link to check how to create an AI agent with IBM Granite directly on Serenity Star.
IBM’s Answer to AI Agent Governance: Automation & Evaluation with watsonx.governance
AI agents are transforming industries, but their autonomy brings risks—ranging from data bias to security concerns. To address these challenges, IBM has introduced new governance and evaluation capabilities within watsonx. Governance.
Why AI Agent Governance Matters
AI agents can operate autonomously, making decisions that impact organizations and customers. Without governance, they can introduce hallucinations, bias, and security risks.
IBM’s governance framework ensures AI agents remain compliant, secure, and aligned with business objectives.
New Agentic Evaluation Metrics
A tech preview of advanced agentic evaluation metrics is now available, helping organizations monitor and refine AI agent performance:
- Context Relevance: Ensures retrieved data aligns with the user’s query.
- Faithfulness: Evaluate whether the generated response stays faithful to the source material.
- Answer Similarity: Measures how closely the agent’s response matches a reference answer.
These metrics help detect early warning signs of errors, ensuring AI agents function responsibly.
Lifecycle Governance for AI Agents
watsonx. governance automates risk, compliance, and security tracking, providing an end-to-end governance framework. A new demo showcases how it helps organizations:
- Define AI use cases
- Associate agents with workflows
- Monitor agent performance in real-time
Upcoming Enhancements in AI Agent Governance
Later this year, IBM will introduce specialized agentic AI metrics, including:
- Query Translation Faithfulness: Ensures agents correctly interpret user queries.
- System Drift Detection: Tracks whether agents evolve safely over time.
- Tool Selection Quality: Verifies that AI agents use the best tools for each task.
With watsonx.governance, businesses can confidently build, deploy, and manage AI agents while maintaining compliance and trust.
Feel free to visit the link for more updates.
Qualcomm and IBM Expand Collaboration to Scale Enterprise AI from Edge to Cloud
Driving the Future of Enterprise-Ready Generative AI
In a groundbreaking move ahead of MWC Barcelona 2025, Qualcomm Technologies, Inc. and IBM have announced an expanded collaboration to deliver enterprise-grade generative AI solutions.
This partnership will enable businesses to leverage AI across edge and cloud environments—enhancing speed, security, governance, and efficiency while reducing costs and energy consumption.
Key Innovations from the Collaboration
🔹 IBM watsonx.governance Meets Qualcomm AI Inference Suite
One of the core advancements in this partnership is the integration of IBM watsonx.governance into the Qualcomm AI Inference Suite. This will help businesses deploy AI responsibly, with strong governance, monitoring, and decision-making capabilities across AI applications.
🔹 Granite 3.1 Models Now Optimized for Qualcomm AI Hub
IBM’s latest Granite 3.1 models—designed for enterprise AI applications—are now optimized for the Qualcomm AI Hub. This gives developers and businesses seamless access to powerful, on-device AI models, ensuring privacy, security, and low-latency performance at the edge.
🔹 Snapdragon and Qualcomm Dragonwing Platforms Power AI at Scale
Qualcomm’s Snapdragon 8 Elite reference design and Qualcomm Dragonwing AI On-Prem Appliance Solution will now support the following:
Granite Guardian 8B and Granite 3.1 8B models for optimized AI inferencing.
- Advanced governance guardrails to monitor and control model operations.
- Enhanced security and privacy features for enterprise deployments.
🔹 Qualcomm Cloud AI Accelerators Now Certified for Red Hat OpenShift
Qualcomm’s Cloud AI family of accelerators has received certification for Red Hat OpenShift, the industry-leading hybrid cloud application platform powered by Kubernetes. This makes it easier for businesses to deploy IBM’s watsonx solutions at scale using Qualcomm Cloud AI hardware.
What This Means for Businesses
This collaboration is set to revolutionize enterprise AI by:
✔ Bringing AI closer to the data—enhancing real-time decision-making at the edge.
✔ Ensuring AI governance and responsibility with watsonx. Governance.
✔ Optimizing performance and efficiency with Qualcomm’s low-power AI platforms.
✔ Simplifying AI deployment with OpenShift-certified cloud AI accelerators.
Businesses now have a robust, scalable AI framework that ensures efficiency, security, and compliance—without compromising performance.
To learn more about this transformative AI collaboration, click here.
OpenAI Expands GPT-4.5 Rollout: More Users Gain Access to Its Most Advanced Model
What’s New with GPT-4.5?
OpenAI has begun rolling out its latest AI model, GPT-4.5, to users subscribed to the ChatGPT Plus tier. This expansion follows its initial launch for ChatGPT Pro ($200/month) subscribers last week. The rollout, expected to take 1-3 days, will adjust rate limits as OpenAI evaluates demand.
GPT-4.5: Bigger, Smarter, and More Expensive
As OpenAI’s largest AI model, GPT-4.5 has been trained on more data and computing power than any previous version. However, its performance is not necessarily the best—on several AI benchmarks, it lags behind newer reasoning models from DeepSeek, Anthropic, and OpenAI.
Additionally, GPT-4.5 comes with high operational costs:
- $75 per million input tokens (~750,000 words)
- $150 per million output tokens
This makes it 30x more expensive than GPT-4o for inputs and 15x more costly for outputs—leading OpenAI to question its long-term availability in the API.
Key Features and Advancements
Despite these challenges, OpenAI claims GPT-4.5 brings several key improvements, including:
1). Deeper world knowledge for more informed responses.
2). Higher emotional intelligence makes interactions more nuanced.
3). Reduced hallucinations, meaning fewer instances of generating incorrect information.
4). Advanced persuasion skills—internal benchmarks show it excels at rhetorical arguments, even convincing another AI to give it cash and reveal secret codes.
What’s Next?
Given its significant computational costs, while GPT-4.5 is being introduced to a broader user base, OpenAI remains uncertain about its future in the API.
Meanwhile, competition in AI reasoning models intensifies, with other players developing more cost-efficient and high-performing alternatives.
To stay updated on OpenAI’s latest developments, visit here.
Stay Ahead with Weekly AI Updates!
AI is evolving rapidly, and missing an update could mean falling behind. Subscribe to NexaQuanta’s weekly newsletter for the latest AI innovations, industry trends, and expert insights. Stay informed, stay ahead!