Sora by OpenAI — Text-to-Video Model, Pricing & Sunset Date

Name: Sora
Rating: 4.5 (800 reviews)
Author: OpenAI

About Sora

Sora is OpenAI's video generation model. It turns a written prompt, or a single reference image, into a short photorealistic clip with camera movement, physically plausible motion and, from Sora 2 onwards, a synchronised soundtrack of dialogue, ambience and effects. It is the model that put AI video in front of a mainstream audience.

OpenAI previewed Sora on 15 February 2024 and opened it to ChatGPT Plus and Pro subscribers on 9 December 2024 as Sora Turbo, generating up to 20 seconds at 1080p for Pro accounts. Sora 2 followed on 30 September 2025 with a standalone social app, markedly better physical realism, native audio and cameos, which dropped a verified likeness into a generated scene. The Videos API arrived a week later and exposed sora-2 and sora-2-pro to developers.

Sora is now being retired. OpenAI announced the wind-down on 24 March 2026, closed the Sora web and mobile apps on 26 April 2026, and will remove the Videos API, including sora-2, sora-2-pro and every dated snapshot, on 24 September 2026. The company said the Sora research team is refocusing on world-simulation work to advance robotics, and that compute had to be traded against products with lower cost per unit of value. No direct replacement has been announced.

Until the shutdown, sora-2 and sora-2-pro still generate 720p to 1080p clips from text or from an image used as the first frame, up to 20 seconds per call, with extensions chaining as many as six continuations into a two-minute sequence. The guardrails are strict: real people including public figures cannot be generated, copyrighted characters and music are rejected, input images containing faces are refused, and output is limited to content suitable for audiences under 18.

If you are building on Sora today, plan the migration now. Runway, Pika, Google's Veo, Kling and Luma's Dream Machine cover the same text-to-video ground, and several of them match the synchronised audio and image conditioning that made Sora 2 distinctive. The endpoint disappears on 24 September 2026 rather than degrading gracefully.

Text to video

A written prompt becomes a shot: framing, camera move, lighting and motion in one pass.

Native synchronised audio

Sora 2 generates dialogue, ambience and effects locked to the picture, not dubbed after.

Image conditioning

Supply a JPEG, PNG or WebP at the target resolution and Sora uses it as the first frame.

20s, extendable to 2 min

Twenty seconds per call, chained through up to six extensions to 120 seconds total.

What Sora Can Do

Text

Understands and writes natural language — prompts in, prose out.

Vision

Reads images: photos, screenshots, diagrams and scanned documents.

Audio

Works with speech and sound as an input or an output, not just text.

Key Features

Text-to-video generation

Describe a scene in natural language and Sora returns a finished clip: subject, camera movement, lighting and motion in a single pass, with no keyframing and no editing timeline.
Synchronised dialogue and sound

Sora 2 was the first OpenAI model to generate a soundtrack with the picture. Spoken lines, room tone, footsteps and effects are timed to the action rather than added afterwards.
Image-to-video

Pass a JPEG, PNG or WebP that matches the target video resolution and the model treats it as the first frame, carrying its composition, palette and subject into the motion.
Extensions up to two minutes

A completed clip can be continued by up to 20 seconds at a time, to a maximum of six extensions or 120 seconds. Extensions do not carry characters or image references forward.
Physical plausibility

Sora 2 holds objects, gravity and contact consistent across a shot, cutting the melting-and-morphing artefacts that gave earlier video models away within a second or two.
Guardrails on every generation

Real people including public figures cannot be generated, copyrighted characters and music are rejected, input images with faces are refused, and output is limited to material suitable for audiences under 18.

Read the documentation

Pricing

Sora is billed per second of finished video, not per token. Standard API rates are below; batch runs at half price. Access ends when the Videos API is removed on 24 September 2026.

Provider	Input	Output	Cached Input	Batch	Fine-Tuning
sora-2 — 720p	—	$0.10 / sec	—	$0.05 / sec	Not available
sora-2-pro — 720p	—	$0.30 / sec	—	$0.15 / sec	Not available
sora-2-pro — 1024p	—	$0.50 / sec	—	$0.25 / sec	Not available
sora-2-pro — 1080p	—	$0.70 / sec	—	$0.35 / sec	Not available

Use Cases

Social and ad creative Vertical 720x1280 clips built for the feed, sized correctly at generation instead of cropped later.
Previsualisation Block out a shot list before a shoot: framing, lens feel and blocking, at storyboard cost.
Course and explainer video Illustrate an abstract idea with a short cutaway instead of another bullet slide.
Product marketing Hero clips and feature loops for landing pages, generated per variant rather than per shoot.
Concept and style tests Try a period, palette, weather or lens before committing a crew and a budget to it.
Localised variants Regenerate the same scene per market instead of re-shooting or re-cutting the edit.
Pitch and animatic filler Fill a deck or an animatic with moving reference in minutes rather than days.
B-roll and cutaways Cover a gap in an edit with a shot you cannot licence, schedule or afford to film.

Reviews

4.5/5

Based on 800 reviews

5 72%

4 18%

3 6%

2 2%

1 2%

No written reviews yet

Be the first to share your experience with Sora.

Share your experience

Log in to write a review for Sora.

Log in to review

Alternatives to Sora

Runway Runway 4.5 Compare Pika Pika Labs 4.3 Compare

Veo Google DeepMind

Kling AI Kuaishou

Dream Machine Luma AI

Stable Video Diffusion Stability AI

Frequently Asked Questions

Is Sora still available?

Only through the API, and only until 24 September 2026. OpenAI closed the Sora web app and the iOS and Android apps on 26 April 2026, so there is no longer a consumer way to generate a Sora clip. Developers can still call sora-2 and sora-2-pro through the OpenAI Videos API until the shutdown date.

When exactly does the Sora API shut down?

24 September 2026. OpenAI notified developers on 24 March 2026 that the Videos API and every Sora 2 alias and snapshot — sora-2, sora-2-pro, sora-2-2025-10-06, sora-2-2025-12-08 and sora-2-pro-2025-10-06 — would be removed on that date. No replacement model has been named.

Why did OpenAI discontinue Sora?

OpenAI said the Sora research team is moving to world-simulation work aimed at robotics, and that it had to make trade-offs on products with high compute costs. The app reached the top of the iPhone App Store shortly after its September 2025 launch but struggled to hold engagement, and the compute behind it was worth more on coding, reasoning and text workloads.

What could Sora actually do?

Turn a text prompt or a single reference image into a video clip of up to 20 seconds, at 720p on sora-2 and up to 1080p on sora-2-pro, with a synchronised soundtrack of dialogue, ambience and effects. A finished clip could be extended by up to 20 seconds at a time, to a maximum of six extensions or 120 seconds in total.

How much does the Sora API cost?

Sora is billed per second of finished video. sora-2 is $0.10 per second at 720p. sora-2-pro is $0.30 per second at 720p, $0.50 at 1024p and $0.70 at 1080p. Batch requests are half price at every tier, so a 10-second 720p clip on sora-2 costs $1.00 standard or $0.50 batched.

Does Sora generate sound as well as picture?

Yes, from Sora 2 onwards. The model generates dialogue, ambience and sound effects together with the image, timed to the action rather than dubbed on afterwards. The original Sora released in 2024 was silent.

Can Sora generate videos of real people?

No. The API rejects real people including public figures, blocks character uploads depicting a human likeness by default, and refuses input images containing faces. Copyrighted characters and copyrighted music are also rejected, and output is limited to content suitable for audiences under 18. The cameo feature that placed a verified likeness into a scene lived in the Sora app and went away with it.

Is Sora open source, and can it be fine-tuned?

Neither. Sora's weights were never released and OpenAI offers no fine-tuning path for it. Prompting, image conditioning and extensions were the only levers available to users.

What should I use instead of Sora?

Runway and Pika are the closest general-purpose replacements listed in this directory. Google's Veo, Kling and Luma's Dream Machine cover the same text-to-video ground, and several now match Sora 2's synchronised audio and image conditioning. If you run Sora in production, migrate before 24 September 2026 — the endpoint is removed on that date rather than degraded.

Sora

At a Glance