ModelsLab

ModelsLab's real product is consolidation, not any single model. You get one API key, one authentication scheme, one billing line and one request shape across 219 models — Flux 3 for 4K text-to-image with bounding-box control, GPT-Image-2.5 Flare and Sunburst for editing up to 14 input images, Qwen Image 2.1, Wan 3.0 for 30-second video with reference-to-video consistency, MiniMax H3 for 15-second clips with native stereo audio, and LTX 2.5 Pro for cinematic 720p/1080p output with synchronized dialogue. Swapping between them is a parameter change, not a rewrite. The developer surface is unusually complete: SDKs for Python, TypeScript, PHP, Dart and Go, a CLI, an MCP server that plugs into Claude Code and Cursor, and an Agents Control Plane where an agent can POST /auth/signup, get a Stripe checkout URL and start a subscription without a human in the loop. The workflow library and workflow builder cover teams that want pre-assembled chains rather than raw calls. The economics are where you need to read carefully. The $21/mo Basic Prototype tier gives 3,250 API calls and 5 concurrent requests; $47/mo Standard Production gives 10,000 calls and 10 concurrent requests; $149/mo Open Source Unlimited removes the call limit on open-source models hosted on ModelsLab's own infrastructure and raises concurrency to 15 parallel requests. That last tier is genuinely flat-rate — but only for open-source models. Google, Kling, Seedance, Vidu and ByteDance models bill per call from a wallet you fund separately, and wallet funding is only unlocked once you hold an active subscription. Video is priced per second of output or as a multiplier on the model's base rate, so a single clip can cost considerably more than an image. Plan allowances reset each billing period and unused calls do not roll over, though wallet balance persists. Every plan reaches every model — tiers differ on volume and concurrency, never on catalog access. Where it fits: product teams shipping media features who need to A/B several models without three vendor contracts, agent builders wiring generation into coding agents, and teams that want unlimited open-source generations at a predictable $149/mo. Where it doesn't: non-technical users who want a visual no-code canvas, single-model shops that would get a lower unit price going direct to ElevenLabs, Kling or Runway, and high-volume premium-video operations that need flat, forecastable clip costs. Concurrency caps at 5, 10 or 15 parallel requests depending on plan, which matters if you run batch jobs. Monthly and yearly plans carry a 100% refund policy and you can cancel any time from Dashboard → Subscription.