If you want every major model behind one key and one bill, use OpenRouter (5.5% fee on credits) or Vercel AI Gateway (list price, zero markup, $5 a month free). If you're in AWS or Google Cloud for compliance reasons, use Amazon Bedrock or Vertex AI. For raw speed on open models, Groq is the fastest host we've used. For image and video models, Replicate and fal let you add a new model in minutes. That's how our internal video studio works.
Gateways charge in three ways: a fee on top of list price (OpenRouter), list price with no markup (Vercel, Bedrock and Vertex pass through each model's price) or per hardware-second or output (Replicate, fal). Model prices themselves are on our LLM API pricing page.
The Ryz Labs video studio from our first episode exists because of this idea:
avoid, like, high subscription fees over a large population of people and instead leverage, like, API keys as a consumption layer so that just drives down the cost of the use case, and you just kinda pay for what you use.
— Jordan Metzner, Episode 1, at 04:15
Adding a model was as simple as copying Replicate's sample API call and schema into the Replit agent (at 24:04). But every model behaves differently, and in the live demo some of them failed:
sometimes I have to go back and forth, make some testing… some of these models failed, so I'm gonna have to go back and, you know, kind of understand why they failed.
— Jordan Metzner, Episode 1, at 24:54
The cost model is simple, and so is the catch. As Sam put it: "every time we generate a video, we're paying replicate" (at 25:28).
Jordan demoed gpt-oss-120b on Groq writing ten marketing emails and called it "incredibly fast". The output wasn't perfect, so he re-ran it with a "no dashes" instruction, at about four to five seconds a run (at 19:00). On a heavier job, a finance dashboard reading P&Ls and balance sheets, the open model "wasn't enough" and he moved to Gemini (at 06:59).
I don't think any any of these LLMs have done a good job of, I would say, like, routing… Amazon has Bedrock which is kind of like this, like, LLM kind of routing platform.
— Jordan Metzner, Episode 23, at 26:47
We have no build of our own with OpenRouter, Vercel AI Gateway, Vertex AI or fal (Jordan named fal only as a Replicate alternative), so they appear in the price table only.
The healthcare company ARYA reaches Anthropic only through Bedrock and Gemini only through Vertex AI, which "make it possible for us to build HIPAA compliant workflows" (at 11:22). The cost of that choice is waiting on the cloud: "I remember Bedrock, like, initially, they didn't have it. Like, we we pushed them to get a particular version" (at 12:40). The same team saw a bill "about six x" what they expected after a change in how long their Lambda held the Anthropic connection open (at 14:33).
OpenRouter passes models through at provider list price and charges a 5.5% fee on credit purchases on pay as you go (8% on Business). Bring-your-own-key use is free for the first $25,000 of list-price inference a month, then 5%. Free models are limited to 20 requests a minute and 50 a day (November 21, 2026).
Vercel's documentation says AI Gateway uses provider list rates with zero markup. The free tier includes $5 of credit a month on free-tier models; paid use runs on purchased credits, and payment processing fees may apply.
Both host many image, video and open models behind one API. Replicate bills public models by hardware time per second (an H100 is $0.001525 a second) or per output, for example $0.04 per Flux 1.1 Pro image. fal bills per output, such as $0.08 per Nano Banana 2 image or per second of video.
Usually for compliance and procurement. A healthcare guest on our show said they only reach Anthropic through Bedrock and Gemini through Vertex AI, because their cloud contracts let them build HIPAA-compliant workflows. The trade-off they mentioned was waiting for new model versions to appear on Bedrock.
No. Groq is an inference company that hosts open models such as gpt-oss on its own chips; Grok is xAI's model family. On November 21, 2026, Groq listed gpt-oss-120b at $0.15 per million input tokens and $0.60 output.