Sora 2 Costs Ten Cents a Second, and the Retries Set the Real Price

Sora 2 Costs Ten Cents a Second, and the Retries Set the Real Price

Activepieces published a breakdown of Sora 2 API pricing per second and per plan, and its one-line thesis is the useful part: “The real price is the sticker price times your attempts.” Sora 2 through the API is $0.10 a second at 720p; Sora 2 Pro at 1080p is $0.70, seven times the rate. Thirty 10-second clips, 300 seconds of footage, is $30 at the bottom tier and $210 at the top, against flat app plans of $20 and $200 a month. Then the multiplier: three attempts per usable clip for a shot list of standalone clips, and 16.67 once scenes have to match, so Pro at three attempts ($21.00 a clip) already costs more than Sora 2 at sixteen.

Read the sources before the arithmetic. The per-tier project totals come from FrameSurfer and the attempt counts from creator reports collected by Invideo, not from OpenAI, and the post’s own plan table marks most cells “not stated,” including the Pro API rate, Batch, and every resolution and watermark column. It is also an Activepieces post with an Activepieces pitch in it: “one credit covers a whole run.” What survives independent of all that is the shape, which is true of any per-second meter: length scales the bill linearly, and every discarded take is billed like the kept one. The advice that follows is sound: draft short at 720p, render only the keepers higher.

The catalog has the API this story is about. The OpenAI provider page lists the OpenAI Videos API among 62 API pages, and its record describes batch inference at half price, where the post’s Batch row says “not stated.” The budgeting controls the post tells you to set up are also in the catalog, as the OpenAI Usage API and the OpenAI Spend alerts API. On the other side of the pitch, the Activepieces provider page lists 26 API pages, including the Activepieces Flow Runs API that a batch of generations would run through.

OpenAI scores 81.8 on the Kin Score, exemplar, with developer ergonomics at 96.4 and operational transparency at 94.7; contract governance at 45.5 is the weak facet. Agent Readiness is 58.4, agent-native, with idempotency, error semantics and rate-limit signal lit and agentic commerce unlit. Activepieces scores 56.3, strong, and 33.0 for agent readiness, with idempotency and error semantics unlit. The catalog can tell you what each call costs to describe. Only your shot list can tell you how many times you will make it.

← Ofox Says Test an OpenAI-Compatible API Before an Agent Touches It, and Its Own Error Semantics Are Unlit