Claude Models Are Live in Momen: Build AI Agents Without the API Plumbing

Nine Claude models landed in Momen's model list at once — Opus, Sonnet, and Haiku. You pick one from a dropdown inside an AI Agent, and it runs against your project with no API key, no SDK, and no middleware in between.

What makes Claude worth wiring in here is not chat. It is tool calling and structured output: a Claude agent in Momen can call an Actionflow, read a table, hit an external API, and hand back a typed object your app can act on. The model reasons about what to do; your backend does it.

Understanding the Claude Models in Momen

They are not a ladder to climb. They are three families, and which family you pick matters more than which version number sits after it.

Opus is the most capable and the most expensive; Anthropic positions it for complex agentic coding and enterprise work. Sonnet sits in the middle and is the sensible default for production traffic. Haiku is the fast, cheap one, for high-volume steps where latency matters more than depth.

Per-model specs are in Anthropic's models overview.

The context window cuts across the families rather than following them. Every 4.5-generation model — Opus 4.5, Sonnet 4.5, and Haiku 4.5 — gives you 200K tokens of context and a 64K output ceiling, while everything from 4.6 up gives you 1M and 128K. If your agent has to read a long document or carry a long transcript, that split decides the model before family or cost does.

Every Claude model here is an inference model: text and images in, text out. None of them generate images. If your app needs image generation, that is a different model in the same list, and a Claude agent can be the step that writes the prompt for it.

Structured output is available on Claude models in Momen, so an agent can return typed fields instead of prose you have to parse.

How to Use Claude in Momen

Momen connects a model two ways: as a built-in model billed in AI Points, or through BYOM (Bring Your Own Model) on your own provider key. Claude is available both ways, and the two paths reach different generations of the model family.

Built-in models

1. Open the AI page in the left toolbar and go to the Agent tab.

2. Click Add agent, or open an agent you already have.

3. Click the model name at the top of the agent editor and choose a Claude model. The same panel holds Temperature, Max number of rounds, Maximum output tokens, and Image processing (Simple or Detailed).

4. Write the System and User halves of the Prompt template. Add Inputs for anything the prompt needs as a variable, and attach Contexts — Data or API — for knowledge the agent should retrieve at run time.

5. Under Tools, give the agent what it is allowed to act on: an Actionflow, an API, or another Agent.

6. In Outputs, choose Plain text or Structured, and switch on Streaming output if your UI should stream tokens.

7. Test it under Debug & preview with Start conversation.

The agent then drops into an Actionflow like any other node, so AI output can trigger a database write, an email, or a payment. The AI integration guide covers that wiring, and the Actionflow node list covers what it can chain into.

BYOM (Bring Your Own Model)

1. On the AI page, open the Model tab and click Add model.

2. Give the model a Name and leave Type on Inference.

3. Set Source to OpenRouter. The Source list also offers OpenAI, Self-hosted, Groq, xAI, La Plateforme, and cohere, but OpenRouter is the only one of them that carries Claude.

4. Pick the model. OpenRouter lists eight Claude entries: Claude-3.7-Sonnet, Claude-3.7-Sonnet-(self-moderated), Claude-3.7 Sonnet-(thinking), Claude-3.5-haiku-20241022, Claude-3.5-sonnet, Claude-3-haiku, Claude-3-opus, and Claude-3-sonnet. The (thinking) entry is the reasoning variant; the (self-moderated) entry is OpenRouter's unmoderated routing of the same model.

5. Paste your OpenRouter key into API KEY and click Verify to move to Verify & Save.

The model then appears in the Model tab with OpenRouter in the Source column, and in the agent model dropdown next to the built-in ones. Capabilities carry over: Claude-3.7-Sonnet added this way supports streaming, image input, tool calling, and structured output.

Two things to know before you choose this path. BYOM is not included on the Free plan; it comes with Basic and Pro. And the Claude models reachable through OpenRouter here are the 3.x generation, while the built-in list carries the 4.5 through 5 generation.

What It Costs

Point conversion

Momen charges built-in models in AI Points, per token, at these rates:

What one agent call costs

The formula is (input tokens × input conversion) + (output tokens × output conversion).

Take a realistic agent call: a 2,000-token input — system prompt, the user's message, and a few rows of retrieved context — and a 500-token answer. On Claude Sonnet 5:

Input: 2,000 × 1.7100 = 3,420 points Output: 500 × 8.5700 = 4,285 points Total ≈ 7,705 points

Run the same call across the family:

Two comparisons fall out of that arithmetic. An Opus call costs about 2.5× a Sonnet 5 call (19,295 ÷ 7,705) and about 5× a Haiku 4.5 call (19,295 ÷ 3,865). And Sonnet 5 is the cheapest Sonnet, not the most expensive: at 7,705 points against 11,570, it runs at about two thirds the cost of Sonnet 4.6 and Sonnet 4.5.

One caveat on the token counts themselves. Claude Opus 4.7 and later — which includes Opus 4.8, Opus 5, and Sonnet 5 — use a newer tokenizer that produces roughly 30% more tokens for the same text than the one Sonnet 4.6 and earlier models use. Points are charged per token, so identical prompts do not produce identical token counts across the two groups. Anthropic documents the change on its pricing page.

How far the Free plan goes

The Free plan includes 100,000 AI Points per month. Against the call above:

Claude Opus 5: 100,000 ÷ 19,295 ≈ 5 calls Claude Sonnet 4.6: 100,000 ÷ 11,570 ≈ 8 calls Claude Sonnet 5: 100,000 ÷ 7,705 ≈ 12 calls Claude Haiku 4.5: 100,000 ÷ 3,865 ≈ 25 calls

That is enough to build and test an agent, not to run one in production. Points top up at 6,000,000 for $10, which puts one point at $10 ÷ 6,000,000. A 7,705-point Sonnet 5 call is about $0.013; the same call on Haiku 4.5 is about $0.006, and on Opus 5 about $0.032.

Plans and top-ups

Basic includes 1,000,000 AI Points per month and Pro includes 5,000,000. On the same 2,000-in / 500-out call, Basic covers roughly 129 Sonnet 5 calls or 258 Haiku 4.5 calls per month, and Pro roughly 648 and 1,293. Plan details are on the pricing page, and the usage calculator will size a specific workload with your own token counts.

Real Showcases

These are apps already running on Momen where a reasoning agent, not a chatbot, is the working part.

A multi-round AI judge. The Momen team built an agent that runs several rounds of dialogue before returning a verdict, with each round's output feeding the next. It is the clearest example of an agent as a multi-step process rather than a single call. Read more

Resume-to-job match scoring. Rajeevdaz built an app that scores a resume against a specific job posting before the user applies. The agent has to produce a consistent, comparable score rather than an opinion, which is exactly what structured output is for. Read more

Landing page critique. MakerThrive shipped a tool that tells founders what is wrong with their landing page. The agent reads a page and returns specific, actionable findings, with the results stored in Momen's database so they can be revisited. Read more

Conclusion

The hard part of shipping an AI feature was never the model call. It is everything the call sits inside: a database that holds the result, an Actionflow that decides what happens next, user accounts so the right person sees it, storage for the files, and payments if you want to charge for it. Momen gives you those as first-class pieces, which is what turns a Claude agent from a demo into a product. Start from the app you actually want to build, and make the agent one step in it.