Launching Prompt Adaptation
Nine months ago we launched the world’s most powerful model router, successfully helping over 100,000 users predictively determine the best suited LLM for every query. Our router outperforms every individual foundation model across all major benchmarks, delivering higher accuracy at lower cost and latency.
But choosing the right model is only half the battle. The next big challenge is prompting each model correctly. As the AI landscape becomes increasingly fragmented, every developer has seen firsthand that prompts written for one model don’t easily transfer to others. Manually rewriting them for every new model that’s released quickly becomes unsustainable, with teams spending thousands of hours on manual prompt engineering as they begin scaling to dozens of different prompts across a constantly shifting pool of models.
Today, we're excited to launch early access for Prompt Adaptation, a state-of-the-art agentic prompt rewriting system that automatically adapts prompts across different models, significantly enhancing accuracy and dramatically cutting down engineering time.
Our enterprise and startup customers have already begun benefitting from Prompt Adaptation. At SAP Sapphire in 2025, Philipp Herzig, Chief Technology Officer and Chief AI Officer of SAP SE, showcased a preview of the SAP prompt optimization service integration into SAP’s generative AI hub powered by Not Diamond with the intent of driving accuracy improvements on its use cases and accelerating engineering throughput:
“Prompt optimization is key to make Enterprise AI more flexible, cheaper, more accurate, and less brittle,” said Philipp Herzig, Chief Technology Officer and Chief AI Officer at SAP. “With Not Diamond, we can adapt to AI innovations and new models much faster and benefit from performance improvements in AI development. This is just the beginning—we’re excited to continue building the multi-model future together.”
Additionally, we're excited to announce additional funding from IBM, SAP.iO Fund, Myriad Venture Partners, Deepwater, DNX, Ambush Capital, and continued support from Defy. This investment underscores the growing importance of prompt adaptation and multi-model orchestration in enterprise AI.
“Generative AI requires a fit-for-purpose model strategy. Smaller, open-source models are how that future will be built, and to harness the power of AI, we need intelligent multi-model infrastructure,” said Emily Fontaine, Vice President, Global Head of Venture Capital at IBM. “Not Diamond combines deep research with real-world impact. Their routing and prompt adaptation systems help scale AI, and we’re thrilled to support their vision.”
State of the Art Performance
Prompt Adaptation outperforms other techniques in the space, including Meta's prompt optimizer powered by DSPy, Amazon Bedrock Prompt Optimization, and Anthropic's prompt improver.
On enterprise client datasets, Prompt Adaptation achieves accuracy improvements of between 5% and 60%. These improvements span various critical use cases, including RAG, data extraction, and text-to-sql. Tasks that previously required up to 40 hours of manual prompt engineering can now be completed in less than an hour, saving thousands of engineering hours annually.
How Prompt Adaptation Works
Prompt Adaptation simplifies prompt engineering into a streamlined, automated workflow:
- Input Collection: Provide your existing prompt, a set of validated input-output examples, your chosen evaluation metrics, and target models.
- Automated Optimization: In the background, Not Diamond iterates over thousands of potential prompts to arrive at the highest-performing prompt for each model.
- Rapid Deployment: Within 30 minutes, you can deploy tailored prompts optimized for performance, cost, and regulatory compliance across your entire model stack.
The Multi-Model Future
The future is multi-model. There won’t be a single giant model everyone uses for everything, but many diverse models and agents. This is why we’re building Not Diamond. We believe this will not only be a better future for AI, but a safer one as well.
Intelligent multi-model infrastructure unlocks model agility for enterprise teams, empowering rapid adoption, optimization, and scalability across an ever-expanding AI model landscape.
Get Early Access
Prompt Adaptation is now available for early access. You can sign up here.