Global investment in artificial intelligence continues to accelerate. Gartner projects that spending on generative AI will climb to $644 billion in 2025, representing a dramatic 77% jump from the previous year. This massive surge highlights how deeply AI is becoming embedded in business strategy. But turning that investment into real value requires clarity—not just enthusiasm—about the models powering today’s AI tools.

Drawing from billions of cloud assets analyzed across industries, the Orca Research Pod has released its latest ranking of the top AI models of 2025. This new list updates last year’s 2024 findings and reflects how organizations are scaling, diversifying, and operationalizing AI.

AI Adoption in the Cloud Keeps Rising

The share of organizations using AI models in their cloud environments has increased dramatically, climbing from 56% in 2024 to 84% in 2025. This reflects a maturing AI landscape where more organizations are moving beyond experimentation and embracing large-scale deployments.

A major driver behind this growth is the widespread use of Azure OpenAI, Microsoft’s managed platform for running OpenAI’s advanced models. The 2025 State of Cloud Security Report found that 30% of organizations now rely on Azure OpenAI, while 27% use Azure Machine Learning (Azure ML) for training, orchestrating, and integrating machine learning workflows.

This mainstream adoption explains why OpenAI models dominate the rankings again this year. It also sheds light on a notable trend: although overall AI use is rising, adoption percentages for individual models have dipped. This isn’t a drop in usage—it’s simply the result of organizations now using a broader mix of models instead of relying heavily on only a few.

What Exactly Is an AI Model?

An AI model is a program trained to recognize patterns and perform tasks by learning from data. Some are built for general-purpose language tasks, while others specialize in vision, speech, coding, embeddings, or reasoning.

AI models usually work in tandem with:

• AI services – cloud-hosted capabilities that allow companies to deploy or customize models at scale
• AI packages – frameworks and libraries used to train, optimize, and operationalize AI

Together, these components form the backbone of the AI-powered applications appearing in modern cloud-native systems.

The 10 Most Used AI Models in 2025

Based on Orca’s analysis of active cloud deployments, these are the models organizations use most frequently. The percentages represent the share of AI-using organizations that have deployed each model.

1. GPT-4o — 44.72%

OpenAI’s flagship GPT-4o (“omni”) leads the rankings. Designed to work seamlessly across text, images, and audio, GPT-4o supports everything from voice-enabled assistants to complex analytical workflows.

Common uses:

• Conversational AI
• Knowledge assistants
• Multilingual support
• Cloud operations copilots
• Security and investigation workflows

2. GPT-3.5 Turbo — 38.20%

Even years after launch, GPT-3.5 Turbo remains a favorite thanks to its balance of speed, accuracy, and cost efficiency. Many production pipelines still rely on it.

Common uses:

• High-volume content generation
• Summaries and translations
• Internal chatbots
• Automated documentation
• Structured text formatting

3. text-embedding-ada-002 — 37.27%

This older embedding model continues to see heavy use, showing how essential semantic search and vector similarity have become in enterprise systems.

Common uses:

• Search indexing
• Recommendation engines
• Clustering
• Deduplication
• Retrieval-augmented generation (RAG)

4. GPT-4o mini — 33.54%

A more lightweight sibling of GPT-4o, this model brings multimodal intelligence at a lower cost and with faster response times—ideal for scaled deployments.

Common uses:

• Real-time chatbots
• Mobile and edge devices
• In-app copilots
• Large-scale customer interactions

5. DALL·E 3 — 23.91%

With its exceptional image-generation capabilities, DALL·E 3 remains the go-to creative model for many teams.

Common uses:

• Marketing visuals
• Product concepts
• UI illustrations
• Creative prototyping
• Visual content A/B testing

6. GPT-4.1 — 22.98%

GPT-4.1 offers improved reasoning and instruction-following, making it suitable for work where accuracy is critical.

Common uses:

• Compliance and policy drafting
• Reports and documentation
• High-assurance analytical workflows

7. GPT-4 — 22.67%

Despite newer releases, GPT-4 continues to play a key role in enterprise environments because of its stability and long-tested behavior.

Common uses:

• Long-form content
• Enterprise chatbots
• Knowledge analysis
• Mature AI applications

8. text-embedding-3-large — 22.36%

A more advanced embedding model that increases recall and semantic precision, especially in technical or large document sets.

Common uses:

• Legal and technical search
• Codebase and policy retrieval
• Knowledge graph enrichment
• Advanced RAG pipelines

9. text-embedding-3-small — 21.74%

A cost-effective cousin of the “large” model, built for high-throughput and latency-sensitive environments.

Common uses:

• Real-time semantic tagging
• Log enrichment
• Personalization at scale
• High-QPS search

10. o3-mini — 19.88%

Although not as widely publicized, o3-mini has quietly gained traction due to its efficiency and low resource requirements.

Common uses:

• Automation workflows
• Lightweight agents
• Operational scripting
• Embedded intelligence in cloud tools

What the Findings Reveal

The Orca Research Pod continuously scans billions of cloud assets and thousands of code repositories to identify emerging patterns in how organizations use AI. This year’s results show:

• Rapid growth in cloud-based AI adoption
• Strong dependence on Azure OpenAI and Azure ML
• Widespread use of OpenAI’s multimodal and embedding models
• Increasing diversification as more models become available

These insights fed into the broader 2025 State of Cloud Security Report, which outlines key risks and security trends in today’s cloud environments.

How Orca Uses AI to Strengthen Cloud Security

Orca became the first CNAPP (Cloud-Native Application Protection Platform) to integrate generative AI back in 2023. Since then, the platform has expanded its capabilities to secure AI-powered environments and to apply AI to cloud security tasks.

Key features include:

• AI Security Posture Management (AI-SPM)
  Tracks risks across 50+ AI models and packages.
• AI-powered guidance and investigations
  Natural-language search, triage assistance, and remediation guidance.
• Multilingual support in 50+ languages
  Making security more accessible for global teams.
• Automated remediation recommendations
  Instantly generated instructions and code.

---

If you’d like, I can also create:

✅ A shorter version
✅ A more formal version
✅ A plagiarism-free SEO blog version
✅ A LinkedIn post version
✅ A Pakistani-audience version
✅ A version tailored to your website

Just tell me!