Is Not Diamond a gateway?

No. Not Diamond says it is not a gateway and that it works with the gateway you already use, which makes the call to the recommended model.

Can I use Not Diamond with Conifer?

Yes. Not Diamond returns the model it recommends, and your code then calls that model with the SDK you choose. That call can go to Conifer, for models that are on Not Diamond’s supported list and in the Conifer catalog.

Does Conifer charge for routing?

No. A request sent to the router costs only the model that answers it, and every response names the model the router chose and what the request cost.

Comparison

Conifer and Not Diamond compared

Not Diamond recommends which AI model should answer each request. Conifer is a hosted gateway with a router built in, so it picks the model and also serves the answer. This page sets out how each one works, using each company’s own published documentation.

Summary

Not Diamond is a model routing service aimed at coding agents. It returns a recommended model, and the call to that model runs through your own gateway or provider accounts. It also trains routers on your own evaluation data and offers prompt optimization. Conifer is a hosted gateway for more than 230 models, each billed at its maker’s list price from a prepaid balance. Its router, in beta, picks from a pool of models you approve and serves the answer in the same request, with no routing fee.

Not Diamond details were read from notdiamond.ai and docs.notdiamond.ai on 7 November 2026. Plans and prices change; check the linked pages before you decide.

Side by side

Topic	Conifer	Not Diamond
What it does	A hosted gateway with a router built in. The router picks a model, and the gateway serves the answer in the same request	Recommends a model for each request. It is not a gateway: the recommendation is carried out by your own gateway or agent harness
How a request is answered	Send `beta-auto` as the model to the gateway, with one Conifer key. The response is the answer from the chosen model	Your code sends the messages and a list of candidate models to Not Diamond, receives the recommended model, then calls it with the SDK and provider account you choose
Models it can choose	Up to 64 models you approve from a catalog of more than 230. With an empty pool, every model your key can call	The candidate models you send with each request, from its list of supported models. A custom router can also include your own fine-tuned models or any inference endpoint
How it decides	It reads what the request needs and weighs it against the cost of each model in the pool. Only models that can hold the whole request are considered	A pre-trained router, or a custom router trained on your evaluation data. For coding agents, its pricing page says it weighs the request, cache state, feedback, session outcomes and past turns to recommend a model and reasoning effort for each step, and uses reinforcement learning to learn from feedback
Cost and quality	You decide which models are in the pool, and the router weighs each request against their cost	Quality first by default, or a preference for cost or for latency. The pre-trained router also takes a blend of cost and quality on a scale from 0 to 10
What you pay	Routing adds no fee: a request sent to `beta-auto` costs only the model that answers it. Every model is billed at its maker’s published list price per token, with no subscription	$0.05 per million tokens routed on pay as you go, with volume discounts on Enterprise. The models themselves are paid to the provider or gateway that serves them
Accounts	One Conifer key and one prepaid balance cover every model in the catalog, with no provider accounts needed	A Not Diamond key for the recommendation, plus your own access to each model it may pick. Pay as you go is by application
What you get back	Every response names the model the router chose and what the request cost, and marks the request as routed	The recommended model and a session ID for tracking and debugging each recommendation
Request content	To choose a model, the router receives the request content; a request that names a model skips the router. The prompt of a standard request is not stored	With privacy-preserving routing turned on, decisions use only derived request metadata and features computed on your side, and Not Diamond does not see the request or its output. The API can also hash message content
Coding agents	The router works on Chat Completions, Responses, legacy Completions and Anthropic Messages, so agents such as Codex can use it. Setup pages cover Claude Code, Codex, Cursor and other tools	Built for long-running coding agent sessions and a range of agent harnesses. Its pre-trained router for code is in early preview
Beyond routing	The gateway itself: fallbacks you list, provider failover, spending limits and a cost receipt on every response	Prompt optimization, which adapts a prompt to work well across different models
Enterprise terms	Team accounts with owner, admin and member roles, per-key model scope and spend caps, set up on request	SOC 2 and ISO 27001, custom SLAs and 24/7 support. Enterprise adds SSO with SAML, admin controls, organization-wide analytics and privacy-preserving deployment

Which one fits

Not Diamond may suit you if: You already run a gateway or hold provider accounts and want only the choice of model, you want a router trained on your own evaluation data, including fine-tuned models or other endpoints, you want to set the balance between cost and quality on each request, or routing decisions must be made without the request content reaching the router.
Conifer may suit you if: You want one service that picks the model and serves it, on one key and one prepaid balance; you want no fee for routing; you need each response to name the model chosen and the exact amount charged; or you want the router limited to models you approve, with spending limits checked before a request runs.

Using Not Diamond with Conifer

Not Diamond returns the model it recommends and leaves the call to you, so that call can go to Conifer.

Recommend, then call: Ask Not Diamond for a model, then send the request to Conifer with the OpenAI SDK.
Model names: Choose candidates that are both on Not Diamond’s supported list and in the Conifer catalog. Conifer also accepts names with the lab in front. A model Conifer does not carry is refused, not replaced.
Or use the Conifer router alone: Send beta-auto as the model and Conifer picks from your routing pool, set in the console, with no separate routing service.

The router reference has examples for curl, the Conifer SDK and the OpenAI SDK, and Replace your provider key covers pointing a client at Conifer.

Conifer Router: Send one model name and Conifer chooses a model for each request from a pool you approve.
What is an LLM router?: How a router picks a model for each request, and what it should tell you afterwards.
Conifer vs OpenRouter: The two gateways side by side, from each company's own documentation.
Conifer vs Requesty: Two gateways with a router in beta, side by side, from each company's own documentation.
All comparisons: Conifer compared with eleven other LLM gateways and routers, from their own documentation.

Questions

Is Conifer an alternative to Not Diamond?: It depends on what you need. Not Diamond recommends a model and leaves the call to your own gateway or provider accounts. Conifer is a hosted gateway with a router built in: it picks a model from a pool you approve and serves the answer in the same request, with the model and cost on the response.
Is Not Diamond a gateway?: No. Not Diamond says it is not a gateway and that it works with the gateway you already use, which makes the call to the recommended model.
Can I use Not Diamond with Conifer?: Yes. Not Diamond returns the model it recommends, and your code then calls that model with the SDK you choose. That call can go to Conifer, for models that are on Not Diamond’s supported list and in the Conifer catalog.
Does Conifer charge for routing?: No. A request sent to the router costs only the model that answers it, and every response names the model the router chose and what the request cost.