Beyond the model count: evaluating a 700+ model AI gateway across six categories
A model catalog is useful only if it helps you ship a real workflow. A chat assistant, an image tool and a video pipeline do not have identical inputs, billing units or integration requirements. This is a product update
A model catalog is useful only if it helps you ship a real workflow. A chat assistant, an image tool and a video pipeline do not have identical inputs, billing units or integration requirements.
This is a product update from ModelFold. The figures below were checked against the live site on November 16, 2026, China Standard Time; catalogs and prices can change.
What the 703 count actually means
ModelFold currently lists 703 model catalog entries from 92 providers, across six categories:
- Chat: 346.
- Video: 151.
- Image: 131.
- Audio: 53.
- 3D: 17.
- Office: 5.
These are catalog entries, not 703 independently trained foundation models. A model family can have separate task variants or versions, including text-to-image and image-to-image entries.
Why we are not claiming βthe world's largestβ
On the same date, OpenRouter's official public models endpoint returned 467 distinct model IDs. Its initially rendered models page showed zero-valued category counters, so those counters were not used as evidence of catalog size.
That does not establish a global ranking. ModelFold's cross-modal catalog entries and OpenRouter's IDs are not perfectly comparable units. Task variants, free variants, visibility and availability can change the totals. Two catalogs also cannot establish the size of every gateway worldwide.
The defensible description is a large, 700+ entry multimodal catalogβnot a verified world record. Evaluate the tasks you actually need, then compare equivalent configurations and billing units.
A workflow is bigger than a chat completion
A product may use chat to turn a brief into structured instructions, images to explore a visual direction, and video or office tools to package the result.
ModelFold offers one account, wallet and API key, with OpenAI-compatible calling options. This does not mean every video, audio or 3D task fits a Chat Completions request. Use the documented endpoint, parameters and lifecycle for each task.
The practical benefit is a shared access and billing layer, rather than pretending all modalities have the same API shape.
Every model now has a detail pageβincluding 3D
The latest update adds detail pages across the catalog, including 3D. Start from the market, inspect a model's capabilities, parameters and user-facing price, then use its online playground entry point.
A public detail page does not mean every generation is free or anonymous. Follow the playground's account, balance and input requirements. This lets you investigate suitability before building an integration around an unfamiliar model.
Try chat before creating an account
The homepage now includes a simplified model chat experience with a 50-credit anonymous trial, shared per IP. This is not an unlimited allowance per visitor and does not make every catalog model free.
The current interface describes a text-only experience, up to 2,000 characters per message and 512 reply tokens, with rate and daily limits. Sign in for the fuller model experience or follow the Agent integration documentation when you are ready.
Compare final charges, not just headline discounts
The live homepage advertises up to 83% savings against the same model's official price. This is a maximum for eligible configurations, not a blanket discount across the catalog. Compare like-for-like resolution, duration, input/output rates and units.
ModelFold is pay-as-you-go, failed jobs are not charged, and credits settle against actual usage with two-decimal precision. Where a task uses a precharge, distinguish that reservation from the final settled charge. The live user quote and final result are the relevant figures.
Before integrating, answer three questions: what is the billing unit, what was the final charge, and what happens to any reservation when a job fails? A token price and a five-second video price are not interchangeable.
Start with one task, not the entire catalog
Try the homepage chat, shortlist a few models for a concrete task, inspect their detail pages and prices, then implement one minimal documented request. Expand to additional modalities once that workflow makes sense.
Sources checked November 16, 2026 CST: ModelFold homepage and market; OpenRouter catalog and official public model list. No paid generation tests were performed for this article.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.