Not Diamond

Not Diamond positions itself as a specialized AI model router, a layer of intelligence that sits between your application and the growing ecosystem of large language models. Its core thesis is straightforward but powerful: no single LLM is optimal for every query, and the overhead of manually selecting and tuning models is unsustainable at scale. Not Diamond aims to solve this by learning from your evaluation data to predict which model will perform best for each input, thereby improving accuracy by up to 25% while cutting costs by as much as 10x. This is not a proxy or a gateway; it is a recommendation engine that tells you which model to call, and you execute the call yourself, keeping your data client-side. For developers and AI engineers building multi-model pipelines, this approach offers a compelling balance of control and optimization.

Where Not Diamond stands out is in its data-driven routing. Unlike simple round-robin or rule-based selection, Not Diamond uses a predictive model trained on a large, cross-domain evaluation dataset. When you upload your own evaluation data, the router can be fine-tuned to your specific use case within minutes. This means the routing decisions become increasingly tailored to your workload—whether you are routing travel planning queries to a model with strong reasoning versus one with creative flair, or choosing between code-specialized and general models for different coding tasks. The accuracy gains are not theoretical; they come from matching the right model to the right input, avoiding the one-size-fits-all compromise. The cost reduction is equally tangible: by routing simple queries to cheaper models, you avoid paying premium prices for tasks that do not require top-tier reasoning.

Not Diamond also supports automatic prompt adaptation, which is crucial when working with multiple models. Different LLMs respond best to different prompt structures, and Not Diamond can learn the optimal prompt for each model per query type. This extends to joint prompt optimization frameworks like DSPy and SAMMO, allowing you to optimize both the model selection and the prompt simultaneously. This is a significant time saver for teams that would otherwise have to manually craft and test prompts across models.

However, the tool is not without its limits. First, it requires integration effort: you need to have existing evaluation data or be willing to collect it. The router's effectiveness is directly tied to the quality and representativeness of that data. Second, Not Diamond is not a proxy—it does not handle API calls or manage rate limits. You still need to orchestrate calls to each LLM yourself, which adds complexity to your pipeline. Third, while the free tier covers up to 100K monthly routing requests, pricing for higher volumes requires custom quotes, which may be a barrier for smaller teams or those with unpredictable usage. The custom pricing model suggests that Not Diamond is targeting serious production deployments, but it also means costs are less transparent.

Who benefits most? Developers who are tired of manually selecting models and tuning prompts for each one will find Not Diamond a relief. AI engineers building cost-sensitive multi-model pipelines can use it to automatically balance accuracy and expense. Data scientists who have evaluation datasets lying around can put them to work, turning static benchmarks into dynamic routing decisions. Machine learning engineers integrating LLMs into existing systems will appreciate the privacy-by-design approach, as no data is sent to Not Diamond during inference.

A practical buyer should think of Not Diamond as a specialized optimization layer rather than a full-service API management platform. It excels when you have multiple models in your stack and want to squeeze out every bit of performance while keeping costs in check. The decision to adopt it should be based on whether you have the evaluation data and engineering bandwidth to set it up, and whether your usage volume justifies the custom pricing. For teams already juggling several LLMs, Not Diamond can be a force multiplier. For those just starting with a single model, it may be overkill. The key is to see it as a tool for maturity—when your multi-model complexity grows beyond manual management, Not Diamond provides a data-driven path forward.