Not Diamond alternatives for model routing
AutoRouter11 min read
Use AutoRouter, the Mac app for Codex and Claude Code, if you want a model router for the coding agents you already use. It picks a model for each task inside Codex and Claude Code, lets both apps share your Claude and ChatGPT plans, and puts 20+ models in both model pickers with no keys, and it also works in Cursor, Claude Desktop, T3 and Terminal. Not Diamond is a routing API for developers building their own apps, plus Not Diamond Code, a router for Claude Code that opens through a waitlist. If you are routing your own app's requests instead, OpenRouter's Auto Router picks a model on a hosted API with no extra fee, RouteLLM is an open-source router you run yourself, Requesty Auto Routing classifies each conversation on a hosted balance, and the Martian Gateway serves 200+ models where you name each one.
AutoRouter wrote this post, so weigh its rows accordingly. Every claim about another product comes from that product's own site, docs or repository, linked where it is used and read on November 17, 2026 unless the sentence says otherwise.
What is Not Diamond?
Not Diamond is a model routing company with two products: a developer API that recommends a model for each request, and Not Diamond Code, a router for Claude Code sessions. Its homepage now calls it an intelligent model router for coding agents and claims 50%+ cost savings and 5%+ accuracy gains (Not Diamond); those are its own figures.
- The routing API reads a request and returns the model best suited to it from a list you give. You then make the model call yourself with your own SDK and keys. The Python and TypeScript SDKs install with
pip install notdiamondornpm install notdiamond, and calls use a Not Diamond API key (quickstart). - Custom routers train on your own evaluation data: inputs, each model's responses and a score for each (router training). A prompt adaptation endpoint rewrites a prompt written for one model so it works on others.
- Not Diamond Code is a local proxy beside Claude Code that picks the model and reasoning effort for each step of a session. Its docs say raw prompts, code, inputs and outputs are not sent to Not Diamond, only routing metadata, and that it is currently available through a gated waitlist (docs).
- Pricing is $0.05 per million tokens routed on Pay as you go, which you apply for through the waitlist, and Enterprise is priced by sales. The pricing page lists no free tier and says the router adds 100 to 150 ms of latency per request on average (pricing).
Not Diamond vs AutoRouter puts the two side by side in one table, with each setup step.
Why look for a Not Diamond alternative?
Most people look elsewhere because they want routing inside the coding agents they already use, today, on the plans they already pay for. Not Diamond is built for a different job: a developer who adds routing to an app. The gaps a Codex or Claude Code user runs into come from that:
- The API recommends and your app calls. Not Diamond returns a model name, and your code still makes each model call with its own provider keys (quickstart).
- Not Diamond Code opens through a waitlist. You request access with a work email, and the team enables your workspace (docs).
- Its Code docs cover Claude Code. They cover Claude Code in the terminal, VS Code, Cursor, JetBrains, Claude Desktop and GitHub Actions, and they don't mention Codex (IDE extensions).
- Credentials are an Anthropic key, a Bedrock key or forwarded mode. Forwarded mode relays the credential Claude Code already signs in with, and the docs don't mention ChatGPT plans (configuration).
None of that is a flaw for an app builder. This post covers routers for your own app and routers for the coding agents you already use, so you can pick by which one you need.
Which Not Diamond alternatives are worth a look?
Five options cover what people use Not Diamond for, each best at one job:
- AutoRouter is best for developers who code in Codex and Claude Code on Claude or ChatGPT plans. It is a Mac app that adds 20+ frontier and open-weight models to both apps' model pickers with no API keys, plus one more choice, AutoRouter, which picks the model for each task after your own rules. Any connected plan works in either app, so your Claude plan runs inside Codex and your ChatGPT plan inside Claude Code. Free costs $0 and connects two subscriptions; Pro is $50 a month with $20 of included usage and connects any number (AutoRouter).
- OpenRouter's Auto Router is best for an app that wants a model picked on a hosted API. You send
openrouter/autoas the model, it picks a model and reasoning effort from the pool your settings allow, and there is no additional fee beyond the selected model's standard rate.allowed_modelsnarrows the pool, andopenrouter/auto-codeuses a policy tuned for coding agents and harnesses (docs). Credit purchases carry a 5.5% fee on the Standard plan (pricing, read November 16, 2026). - RouteLLM is best for teams that want an open-source router on their own servers. The LMSYS project is Apache 2.0 licensed, installs with
pip install "routellm[serve,eval]", and serves an OpenAI-compatible endpoint where the model name picks a router and threshold, such asrouter-mf-0.5. It routes between one strong and one weak model (GitHub), and its authors reported cutting cost by over 85% on MT Bench against GPT-4 alone while keeping 95% of its performance (LMSYS, July 1, 2024). - Requesty Auto Routing is best for Claude Code on one hosted balance with routes you describe. You set the model to a policy such as
policy/claude-auto-coder, a classifier model you choose reads the first user message of each conversation, and the chosen model is kept for the rest of it. The feature is in beta, and each decision is billed as a normal request to the classifier, once per conversation (docs); Pay as you go adds 5% to model prices (pricing, read November 16, 2026). - The Martian Gateway is best for one key across 200+ models with OpenAI- and Anthropic-compatible APIs. You choose the model for each request, and its docs describe no automatic model picker; they have setup pages for Claude Code, Codex, Cursor, Cline and OpenCode (gateway, Claude Code). Its docs don't list prices.
How do the Not Diamond alternatives compare?
They differ most in who picks the model, who makes the call and what pays for it, as of November 17, 2026:
| Tool | Best for | Who picks the model | Runs as | Price |
|---|---|---|---|---|
| AutoRouter | Codex and Claude Code on plans | AutoRouter, after your rules | Mac app | Free $0; Pro $50/month with $20 of usage; paid models at provider price + 10% |
| Not Diamond | Routing inside your own app | Not Diamond recommends; your app calls | Hosted API; local proxy for Claude Code | $0.05 per million tokens routed, through the waitlist |
| OpenRouter's Auto Router | A hosted pick for apps | OpenRouter, from your allowed models | Hosted API | No router fee; 5.5% fee on credits (Standard) |
| RouteLLM | Self-hosted, open source | A router you train or download | Your servers | Free (Apache 2.0) |
| Requesty Auto Routing | Claude Code on one balance | A classifier you choose, once per conversation | Hosted API | Classifier request per conversation; 5% markup |
| Martian Gateway | One key for 200+ models | You, for each request | Hosted API | Not listed in its docs |
What is an LLM router? explains the kinds of router behind these tools: learned routers like Not Diamond and RouteLLM, classifier routes like Requesty's, gateways like Martian's, and plan-aware routers like AutoRouter.
Which alternative works inside Codex and Claude Code?
AutoRouter works inside both apps with no waitlist, keys or base URLs. Setup takes three steps:
- Download AutoRouter for macOS from autorouter.com. It is signed and notarized by Apple.
- Open it, sign in or create a free account, sign in to your plans, allow Accessibility, and click Enable for each app.
- Start a new session and choose AutoRouter as the model: in Codex, from the model menu under the message box; in Claude Code, with
/model.
From then on, AutoRouter applies your plain-English rules first, such as "Use Claude Opus 5.5 for UI work." Otherwise it works out how hard each task is and only considers models that clear the quality bar for it: hard work goes to the strongest model your plans cover, and quick work to the one that costs you least, often your plan or a low-cost open model. Every request shows its model, what it cost and why. How AutoRouter picks a model walks through each step.
AutoRouter also lets the two apps share your plans. Connect Claude, ChatGPT, Grok and Gemini plans, and any of them can run in either app (how that works). When one account reaches its limit, AutoRouter moves on to the account with the most room left before it spends any credit. AutoRouter Free connects two subscriptions, and Pro connects any number.
The hosted options connect by pointing Claude Code at their API. Requesty's guide sets ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN and ANTHROPIC_MODEL=policy/claude-auto-coder (docs), Martian's sets the base URL to https://api.withmartian.com with your Martian key (docs), and OpenRouter has its own Claude Code guide that bills your OpenRouter credits (guide). Each bills its own balance or your keys instead of your plans. Model routers for Claude Code and Codex, compared lines up more of them.
When is Not Diamond still the better choice?
Not Diamond is the better choice when you route requests inside your own product, want a router trained on your own evaluation data, or run Claude Code on Linux, Windows or in CI with your own Anthropic or Bedrock keys. Its proxy runs on macOS, Linux and Windows and in GitHub Actions (GitHub Actions), its docs say prompts and code go straight to your provider or gateway, and it lists SOC 2 and ISO 27001 (Not Diamond).
AutoRouter runs only on macOS 14 or later, doesn't serve your own apps' API traffic and doesn't train a router on your evaluation data. It sometimes misjudges how hard a task is, and you can rate the pick when it does. Its requests pass through AutoRouter's servers on their way to each provider; it doesn't store transcripts, but keeps the message that opens each request, up to 4,000 characters, to review routing (AutoRouter privacy). For other gateways, OpenRouter alternatives for coding agents covers LiteLLM, Portkey, Vercel AI Gateway and more.
Questions
Is Not Diamond free?
No. Its pricing page lists no free tier: Pay as you go is $0.05 per million tokens routed, with access through the waitlist, and Enterprise pricing comes from sales (pricing). Model calls run on your own provider credentials. AutoRouter Free costs $0 a month and routes across two of your subscriptions.
Does Not Diamond work with Codex?
Its docs don't mention Codex. Not Diamond Code covers Claude Code in the terminal, VS Code, Cursor, JetBrains, Claude Desktop and GitHub Actions (docs). AutoRouter works in Codex (the app and the CLI), the Claude Code CLI, Cursor, Claude Desktop, T3 and Terminal.
Is OpenRouter's Auto Router the same as AutoRouter?
No. OpenRouter's Auto Router is the openrouter/auto model on OpenRouter's API, which picks a model for you and bills your OpenRouter credits (docs). AutoRouter is a separate Mac app for Codex and Claude Code.
Is Not Diamond related to diamond alternatives like moissanite?
No. Not Diamond, at notdiamond.ai, is an AI model routing company. Searches for diamond alternatives mostly return gemstones such as moissanite and lab-grown diamonds, which have nothing to do with it.
Does AutoRouter run on Windows or Linux?
No. AutoRouter runs on macOS 14 or later, and Claude Code needs Node 22.18 or later. On Windows or Linux, Not Diamond Code's proxy and RouteLLM both run, and the hosted routers work from any system.
Sources
- AutoRouter
- AutoRouter privacy
- Not Diamond, read November 17, 2026
- Not Diamond pricing, read November 17, 2026
- Not Diamond routing quickstart, read November 17, 2026
- Not Diamond router training quickstart, read November 17, 2026
- Not Diamond Code docs, read November 17, 2026
- Not Diamond Code configuration, read November 17, 2026
- Not Diamond Code IDE extensions, read November 17, 2026
- Not Diamond Code GitHub Actions, read November 17, 2026
- OpenRouter: Auto Router, read November 17, 2026
- OpenRouter: Claude Code integration, read November 16, 2026
- OpenRouter pricing, read November 16, 2026
- RouteLLM on GitHub, read November 17, 2026
- LMSYS: RouteLLM, July 1, 2024
- Requesty: Auto Routing, read November 17, 2026
- Requesty pricing, read November 16, 2026
- Martian Gateway, read November 17, 2026
- Martian: Claude Code, read November 17, 2026