Sume vs Replicate: running models by API or running a video agent

Replicate and Sume both give developers an API for generating media, at different layers. Replicate runs thousands of community models behind one prediction API and bills by hardware time or by output. Sume is a video agent platform: a video model catalog, plus an agent that plans whole videos, Formats your backend calls, Avatar 1.0, and timeline assembly, billed from one wallet.

Replicate facts come from Replicate's own site and docs, read on 2026-09-24. Sume facts come from Sume's docs, pricing page, and public catalog.

What is Replicate?

Replicate's homepage says: “Run AI with an API.” and “Run and fine-tune models. Deploy custom models. All with one line of code.” It lists “Thousands of models contributed by our community,” and its enterprise page says “Enterprise users get access to 50,000+ models through one API and one contract.”

• API: POST https://api.replicate.com/v1/predictions (HTTP API). “The default behavior of the API is to use async mode,” and a Prefer: wait header turns on sync mode (create a prediction).
• Webhooks: Replicate “will send HTTP POST requests to that URL when the prediction is created, updated, and finished” (webhooks).
• SDKs: official clients for Node.js, Python, Swift, and Go (client libraries).
• Coding agents: an MCP server that “supports all of the operations in Replicate's HTTP API” (MCP), an official CLI, and Cog for packaging models (open source).
• Dedicated endpoints: “Deployments … provide private, dedicated API endpoints” (deployments).

What is Sume?

Sume is a video agent platform: people brief an agent in the Agents tab, save the result as a reusable Format, and call that Format, the agent, or individual models from their own backend over one API, a hosted MCP server, a CLI, and a TypeScript SDK, all billed from one wallet.

• Agents: a chat workspace at sume.com/agents where the agent plans a video, asks before it spends, and generates stills, avatars, voice, lip-sync, and a finished timeline.
• Formats: a saved recipe your backend calls with POST /v1/formats/{handle}/{slug}/runs, plus 27 ready-made Formats at the sume handle. See What is a Sume Format?.
• Models: an OpenRouter-compatible POST /v1/videos endpoint over a catalog of video models, with sume/auto routing, plus image, music, speech, and Avatar 1.0.
• Avatar 1.0: create a reusable avatar ($0.95 per avatar) and generate 4–60 second talking videos at $0.184/s standard, $0.245/s plus, $0.55/s max (no product image).
• Developer surfaces: the Developer API, a hosted MCP server for Claude Code, Cursor, and Codex, a CLI, and a TypeScript SDK.
• Plans are Free $0, Pro $40 / month, Startup $120 / month, Scale $400 / month. Usage draws from one prepaid wallet at each model's published USD rate plus a 5.5% agent fee by default; routed video models are priced at the provider's list price × 1.25. The public catalog states both.

How do Sume and Replicate compare?

What does video cost on Replicate?

Replicate says “Some models are billed by hardware and time, others by input and output” (pricing), and “To use Replicate, you must first purchase credit upfront” (prepaid credit). Published prices for three video models:

• On Sume, routed video models are billed at the provider's list price × 1.25 plus the agent fee, and the plan ladder is Free $0, Pro $40 / month, Startup $120 / month, Scale $400 / month. Exact per-model rates are in the catalog and on API pricing.

Which one fits which job?

• Replicate fits when you want to run or fine-tune a specific model, deploy your own model with Cog, or get a dedicated endpoint, across a very large catalog.
• Sume fits when the deliverable is a finished video rather than one model output: its agent plans the shots, voice, avatar, and timeline, and a saved Format turns that into one API call with an idempotency key, a spend cap, and an optional webhook.
• Both run an MCP server that coding agents can connect to.

How do I try Sume?

Start a thread in the Agents tab, or create an API key and follow the quick start. Plans and per-model rates are on Pricing and API pricing; How Sume pricing works explains the wallet.

Vendor details change often. Everything we say about other products here comes from their own pages as of 2026-09-24; check those pages before you decide.