Cycling through all six. Tap any point to stop.

Measured on six things

One API for every AI model — image, video, audio, 3D, and...

Speed

Lighthouse mobile 69/100 with LCP 5.6s and FCP 4.1s — slow on phones. Desktop 96/100 with CLS 0 and TBT 128ms. The embedded live playground and catalog likely weigh down mobile first paint.

Screenshots

1 / 4

ModelRunner

About this app

ModelRunner is a unified API platform that lets developers access hundreds of AI models across image, video, audio, 3D, and text generation through a single API key. It eliminates the need to manage multiple provider accounts, SDKs, and billing systems. The platform offers a playground for testing models with real example prompts, a curated model catalog, and an MCP server that turns your AI assistant into a model-calling tool. It also features Wrappers, allowing you to package any model into a custom endpoint with preset prompts and exposed inputs, all without building a backend. With per-request pricing, asset management, and typed SDKs for Python and TypeScript, ModelRunner streamlines AI integration for developers and teams.

Key features

1Unified API across image, video, audio, 3D, and text models with one key and per-request pricing.
2In-browser playground with real example prompts and outputs for every model before writing code.
3MCP server with OAuth that turns the catalog into callable tools for Claude, Cursor, and other assistants.
4Wrappers to package any model into a custom endpoint with preset prompts and exposed inputs, no backend needed.
5Typed SDKs for Python and TypeScript, plus an OpenAI-compatible chat endpoint for LLMs.

Target customer profile

Drawn from the product itself, not from a survey.

Demographic

AI/ML engineers and developers

Pain points

Managing multiple AI provider accounts, SDKs, and billing systems; complex integration code; lack of unified testing environment.

Primary needs

Single API for all models, easy model switching, playground for testing, typed SDKs, and MCP server for assistant integration.

Demographic

Product teams and startups

Pain points

Limited backend resources; need to quickly prototype and ship AI features without infrastructure overhead.

Primary needs

Wrappers to create custom endpoints without backend, hosted outputs, asset library, and per-request pricing to control costs.

Demographic

Creative technologists and agencies

Pain points

Juggling various AI tools for image, video, and audio generation; time-consuming asset management and iteration.

Primary needs

Curated collections of production-ready models, seamless asset reuse, and multi-modal generation from one platform.

Scout Review

Written by AI from measured evidence, scored out of 100.

ModelRunner is a unified API platform for running hundreds of AI models across image, video, audio, 3D, and text through one key. The playground is the standout feature — run any model in-browser with real prompts before writing code. MCP server integration with OAuth is a genuine differentiator. Pricing is transparent per-request with no monthly fees. The catalog is well-organized with collections by use case. Weaknesses: mobile performance lags at 69/100 with 5.6s LCP, no CSP header or SPF record, no SOC 2 or compliance claims, and SDK documentation is inconsistent between marketing and docs. The product is competent and clearly positioned but faces stiff competition from established players.

Measure by measure

Usability

Playground lets you run any model in-browser with real prompts before writing code. Docs split into Guides and API reference with an llms.txt index. MCP setup is one npx command with OAuth. Typed SDKs claimed; docs mention JavaScript SDK.

Design

Dark, modern developer aesthetic with a live playground embedded above the fold. Real example prompts, outputs, and per-request pricing sit inline on model pages. Collections group models by use case. Polished, but not visually distinctive versus peers.

Speed

Lighthouse mobile 69/100 with LCP 5.6s and FCP 4.1s — slow on phones. Desktop 96/100 with CLS 0 and TBT 128ms. The embedded live playground and catalog likely weigh down mobile first paint.

Security

HTTPS enforced with HSTS and most security headers present. No CSP header and no SPF record detected. MCP uses OAuth instead of pasted keys. No SOC 2, ISO, or pen-test claims found in fetched content.

Accessibility

Lighthouse accessibility 96/100 with two failing audits: insufficient color contrast and visible text labels not matching accessible names. No WCAG statement or multi-language support mentioned. Responsive layout confirmed by mobile run.

Growth

Real problem: developers juggling provider accounts, SDKs, and billing. Unified API across five modalities plus MCP server and Wrappers give concrete differentiators. Category is crowded but headroom exists for the multi-modal wedge.

Improvement areas

Mobile performance: LCP 5.6s and FCP 4.1s drag the mobile Lighthouse score to 69/100.
Security headers: no Content-Security-Policy and no SPF record detected; no SOC 2 or compliance claims.
Accessibility: two failing audits (color contrast, accessible name mismatch) and no WCAG statement.
Documentation consistency: docs say JavaScript SDK while marketing promises Python and TypeScript.

Design & Usability

ModelRunner's design is competent developer-tool work: dark theme, live playground above the fold, real example prompts and outputs on every model page. The catalog is browsable and collections group models by use case. Usability is where it gets interesting — the playground lets you run any model in-browser before writing code, which is genuinely useful for evaluating whether a model fits. Docs split into Guides and API reference with an llms.txt index for AI assistants. MCP setup is a single npx command with OAuth, no API key pasting. The gap: docs mention JavaScript SDK while marketing promises Python and TypeScript, and there's no clear onboarding walkthrough for first-time users beyond the playground.

Speed & Security

Speed is split: desktop Lighthouse 96/100 with CLS 0 and TBT 128ms, but mobile drops to 69/100 with LCP 5.6s and FCP 4.1s. The embedded live playground and catalog likely weigh down mobile first paint. Security is solid but not exceptional: HTTPS enforced with HSTS, frame protection, X-Content-Type-Options, Referrer-Policy, and Permissions-Policy all present. Missing: Content-Security-Policy header and SPF record. The MCP OAuth flow is a genuine security win over pasted API keys. No SOC 2, ISO, or pen-test claims found, which is normal for a product at this stage but leaves enterprise buyers without a compliance story.

Accessibility & Growth

Accessibility scores 96/100 on Lighthouse with two failing audits: insufficient color contrast and visible text labels not matching accessible names. No WCAG conformance statement or multi-language support mentioned. The site is responsive and keyboard-navigable per automated checks. Growth potential is real but not exceptional: the problem — developers juggling multiple provider accounts, SDKs, and billing — is genuine, and the unified API across image, video, audio, 3D, and text is a concrete differentiator. The MCP server turning the catalog into callable tools for Claude and Cursor is the sharpest wedge. Wrappers for custom endpoints without a backend is a sleeper feature. But the category is crowded with Fal, Replicate, and OpenRouter, and 'unified API' is table stakes language at this point.

Conclusion

ModelRunner is worth a look if you're tired of managing five provider accounts and want one API key for everything from Nano Banana images to Veo video to ElevenLabs audio. The playground alone justifies a free signup — being able to test a model before committing to integration code is genuinely useful. The MCP server is the feature to watch: if you use Claude or Cursor daily, turning the catalog into callable tools changes your workflow. But go in with eyes open: mobile performance needs work, security headers are incomplete, and the compliance story is nonexistent. For indie developers and small teams, it's a solid bet. For enterprise, wait for the SOC 2.

Suggestions for improvement

Add a Content-Security-Policy header and SPF record to complete the security posture.
Fix mobile LCP (5.6s) by lazy-loading the embedded playground and catalog below the fold.
Resolve the SDK documentation inconsistency — docs say JavaScript only, marketing promises Python and TypeScript.
Publish a WCAG conformance statement and fix the two failing accessibility audits (contrast, label names).

How we score

Alternatives comparison

Named competitors, point by point. Nobody paid to appear here or to be left out.

	ModelRunner	Fal	Replicate	Higgsfield	OpenRouter
Unified API across modalities	Image, video, audio, 3D, and text models through one API key and request shape.	Fast inference focused primarily on image and video generation.	Run open-source models via API with a community-driven library.	Specialized in AI video generation from text and images.	Unified API primarily for text and chat LLMs.
In-browser playground with real example prompts	Every model page has a playground with real prompts and outputs to run before coding.	Playground available for testing models with sample inputs.	Interactive demo on model pages with example inputs.	Web-based video generation interface with prompt input.	Chat playground for testing text models side by side.
MCP server for AI assistant integration	MCP server with OAuth turns the catalog into callable tools for Claude, Cursor, and others.	No native MCP server; API integration required.	No native MCP server; community wrappers may exist.	No MCP server; web interface focused.	No native MCP server; OpenAI-compatible API only.
Custom endpoint creation without backend	Wrappers package any model into a custom endpoint with preset prompts and exposed inputs.	No equivalent wrapper feature; direct model API calls.	Deployments allow custom model hosting but require more setup.	No custom endpoint feature; fixed video generation workflows.	No custom endpoint feature; routing and fallback only.
Pricing model	Per-request pricing with no monthly fees; exact cost shown per configuration.	Pay-per-use pricing for inference.	Pay-per-second or per-output depending on model.	Subscription-based pricing for video generation.	Pay-per-token with provider passthrough pricing.
Typed SDKs	Typed SDKs claimed for Python and TypeScript; docs mention JavaScript SDK.	Python and JavaScript client libraries available.	Python and JavaScript SDKs with typed interfaces.	API access available; SDK support limited.	OpenAI-compatible SDKs work; no dedicated typed SDK.