Why IBM invested in Not Diamond: The future of AI is multi-model with a fit-for-purpose AI strategy

With new major models constantly released on the LLM leaderboard, the race to improve performance and efficiency at the model layer continues. Enterprises and startups alike are struggling to keep up.

AI is no longer a game of size, where bigger models mean better outcomes. Startups, such as Not Diamond, are making AI more practical and widespread by developing cost-effective solutions that embrace efficiency. By leveraging an open, multi-model and fit-for-purpose strategy, AI technology can scale in production and become truly transformative. However, we continue to see enterprises struggling, as they face a series of challenging questions:

• Which model should I choose for my application?
• How much should I spend?
• Should I update my existing model selection?
• Should I incorporate reasoning models?
• Should I train on my own data?
• Which prompts work best for my existing bench of approved models?

These are the questions IBM is helping clients answer through innovative solutions.

The future is not a single, monolithic model. It is a network of models—modular, efficient and intelligently orchestrated—driving productivity and growth.

No single model is best at everything—some are faster, cheaper or better for specific domains and regulatory requirements. The idea that one giant model can solve every task is increasingly impractical as performance, cost and compliance needs diverge. Open-source models, such as LLaMA and Mistral, are gaining traction.

Meanwhile, enterprise-focused models are emerging for specialized use cases and regulatory pressures are driving demand for data sovereignty and control. Companies need the flexibility to adapt as the ecosystem evolves—whether it’s for summarization, code generation or compliance.

The solution is not to pick a single model, but to route each prompt to the right model for the task. The CEO of Not Diamond put it perfectly: the world has already realized that a network of computers is better than one enormous computer. The same will be true for models: a network of models will outcompete one large model.

IBM views Not Diamond as building essential infrastructure for the next chapter of enterprise AI, which aligns with our vision of open, secure and flexible AI ecosystems. Just as Kubernetes became the orchestration layer for cloud-native applications, model routers like Not Diamond can be the backbone of AI-native systems.

Not Diamond is building a platform that instantly routes queries across models, dynamically adapts prompts, and lets enterprises adopt new models without delays, rewrites or lock-in. Their prompt adaptation system automatically rewrites prompts for different models, improving accuracy by up to 60% and cutting manual prompt engineering time from hours to minutes.

Teams are no longer required to adjust prompts for every new model manually. Prompt adaptation searches thousands of variations and deploys the most efficient and effective solution.

Along with model routing, Not Diamond’s platform is a critical enabler of a fit-for-purpose AI strategy: the best prompts for your models, and the best models for your customers, seamlessly integrated into enterprise workflows.

IBM is proud to support Not Diamond as we collectively build a multi-model strategy.