Not Diamond

Not Diamond

Intelligent model routing for coding agents, designed to fit existing harnesses and gateways.

paid
Last Updated 8/28/2026

About Not Diamond

Not Diamond is an intelligent model router built for coding-agent workloads. It evaluates each step in a session and recommends a model and reasoning effort from the available pool, using signals such as payload semantics, cache state, prior turns, sub-agent structure, compaction events, and session outcomes. The router integrates with an existing agent harness and gateway instead of replacing the gateway. It suits engineering teams that want to control inference spend without defaulting every task to the most expensive model. The pay-as-you-go rate is $0.05 per million tokens routed, while enterprise pricing is customized and includes volume discounts, organization-wide analytics, security and admin controls, SAML SSO, SLAs, and around-the-clock support. Routing adds an average 100–200 ms per step. Privacy-preserving mode bases decisions on derived request characteristics and locally computed features rather than payloads, inputs, or outputs. The service is SOC 2 and ISO 27001 compliant. It works with multiple coding-agent harnesses, including Claude Code, and continuously learns from usage feedback and session results. The tradeoff is straightforward: teams add a routing dependency and some latency in exchange for model selection tuned to cost and quality.

Best for

Engineering teams running coding agents across multiple models and seeking lower inference costs.
Platform teams that already operate an AI gateway and want a separate model-selection layer.
Organizations that need SAML SSO, administrative controls, analytics, and service-level support.

Best use cases

A coding-agent session needs a model recommendation for each step based on cost and quality objectives.

Inference spending is rising because every request defaults to the strongest and most expensive model.

An existing harness and gateway need intelligent routing without replacing the current access layer.

Compare before choosing

Unlike an AI gateway, Not Diamond recommends a model while the existing gateway manages access.

Pay-as-you-go pricing charges a fixed routing fee instead of bundling the underlying model usage.

Privacy-preserving routing can make decisions from derived request characteristics and locally computed features without reading payloads.

Features

Route coding-agent steps

Select a model and reasoning effort for each step in a long-running coding-agent session.

Reduce inference spend

Move suitable requests away from a default frontier model while preserving a defined quality objective.

Add routing to existing infrastructure

Integrate intelligent model selection with an existing agent harness and AI gateway.

Pros

  • The $0.05 per million routed tokens rate gives small teams a clear routing cost.
  • It integrates with existing harnesses and gateways, reducing infrastructure replacement work.
  • Privacy-preserving routing avoids sending payloads, inputs, or outputs to the routing service.

Cons

  • Each routing recommendation adds an average 100–200 milliseconds to an agent step.
  • The service becomes another dependency in the coding-agent execution path.
  • Enterprise pricing requires a sales conversation, so its total cost is not immediately calculable.

Pricing Plans

Pay-as-you-go

$0.05 per million tokens routedusage-based
  • Multi-harness support
  • Savings and usage dashboard
  • Continuous learning from harness usage
  • Privacy-preserving routing

Enterprise

Custom pricingcustom
  • Volume-based discounts
  • Organization-wide analytics
  • Advanced security and admin controls
  • SAML SSO
  • SLAs and 24/7 support

* Pricing information is for reference only. Please refer to the latest announcement on Not Diamond's official website.

FAQ

It predicts which model to use for each prompt so a team can pursue the best quality available within its cost objective.
No. It recommends which model to use and integrates with existing gateway infrastructure, while the gateway continues to manage model access.
Not Diamond reports an average added latency of 100–200 milliseconds for each routing recommendation.

Tool Information

DeveloperNot Diamond
Release Date4/5/2026
Pricingpaid