Not Diamond

Not Diamond: Route Every Query to the Right Model

Not Diamond is an intelligent model router: a layer that sits in front of your LLM calls and predicts, per input, which model will produce the strongest answer at the lowest cost. Instead of hard-coding one frontier model for everything, a coding agent or application routed through Not Diamond sends easy requests to cheap models and hard ones to frontier models — automatically, based on routing models trained on evaluation data.

It is built for engineering teams running coding agents and high-volume LLM workloads. Not Diamond is stack-agnostic: it integrates with existing harnesses and gateways rather than replacing them, so there is no vendor lock-in on the inference side. Teams define quality and cost preferences, and the router learns continuously from production traffic while preserving privacy.

Key Features

• Per-query routing across multiple LLMs, tuned to maximize response quality while minimizing cost
• Reported results of roughly 5% accuracy gains and 30% cost savings, with typical routing overhead of 100-200ms per step
• Stack-agnostic integration that layers onto existing agent harnesses and LLM gateways
• Continuous learning: routing improves from your own evaluation and production data
• Pay-as-you-go pricing at $0.05 per million tokens routed, plus an enterprise tier with volume discounts, SSO/SAML, and SLAs
• SOC-2 and ISO 27001 compliance with privacy-preserving routing

Ideal Use Case

An engineering team running coding agents at scale whose inference bill is dominated by a single frontier model: drop Not Diamond in as the routing layer, keep the existing gateway, and let easy requests flow to cheaper models while hard ones still reach the frontier — recovering 20-40% of spend without a measurable quality drop.

How Not Diamond differentiates

Not Diamond routes rather than resells inference, so it complements gateways instead of competing on margin. Its production credibility is unusual for its size: OpenRouter's CEO publicly credits it with reducing inference costs while improving output quality, Rootly reported a 39% accuracy increase across SRE benchmarks, and its customer logos include Hugging Face, DoorDash, IBM, and American Express. The company raised a $2.3M seed from defy.vc with angels including Google DeepMind's Jeff Dean, Databricks chairman Ion Stoica, and GitHub co-founder Tom Preston-Werner.

FAQ

What is Not Diamond? Not Diamond is a model routing layer that predicts the most suitable LLM for each query, balancing quality and cost across providers.

How much does Not Diamond cost? Pay-as-you-go pricing is $0.05 per million tokens routed; enterprise plans add volume discounts, SSO, and SLAs. There is no free tier.

Does it replace my LLM gateway? No — it is stack-agnostic and layers onto existing harnesses and gateways such as OpenRouter.

How much can routing save? Not Diamond reports teams typically save 20-40% on inference without quality degradation, with 100-200ms added latency per routed step.

tl;dr

Not Diamond is a per-query LLM router used by OpenRouter, Rootly, and others to cut inference costs 20-40% while holding or improving quality — priced at $0.05 per million tokens routed.